Join us as we learn strategies for acknowledging, valuing, and respecting differences in the workplace and leave with action steps for building communication around diversity, equity, and inclusion. Lunch will be served.
Event: Diversity Activity and Q&A with Cedric Stallworth
Date: Tuesday, October 18
Time: 11 a.m. - 12:30 p.m.
Place: Klaus Bldg. Room 1116W
To send your RSVP and submit your question for the Q&A session with Cedric Stallworth, visit https://gatech.co1.qualtrics.com/jfe/form/SV_cDaZCUdpLuJ6DPw.
]]>Where: GT Hotel Club Room
800 Spring St NW, Atlanta, GA 30308
the Club Room is located on the first floor of the Hotel
When: Thursday, March 31, 5 - 8 p.m.
There are two pool tables in the club room and we have reserved a photo booth so you can have a souvenir of the evening. Refreshments and drink tickets provided. So take a break from your studies, and come relax and have fun!
March 11
11 a.m. - 2 p.m.
In-person and virtual participation welcome. Register here to save your spot! https://tinyurl.com/WomenInDataScience22
The College of Computing and the Machine Learning Center at Georgia Tech are hosting Dean, who is speaking on March 8, in the Clough Auditorium (144). His presentation, Five Exciting Trends in Machine Learning, begins at 11 a.m.
]]>Power of Two Schedule
]]>9 -10:30 a.m.| Breakfast for everyone in Computing!... bacon - Klaus Atrium
11 a.m. -12 p.m. | Computing Student Conversations w/ the Dean - Klaus 1443
1200pm-130pm | GT Computing Faculty and Staff Lunch - Klaus Atrium
2 -4 p.m. | Afternoon snacks and games for everyone in Computing - Klaus 1116
2:22 p.m. | Giveaways - Klaus 1116
5:00 p.m. | In-person Fireside Chat with Dean Isbell - Clough 144
(Join virtually https://primetime.bluejeans.com/a2m/live-event/fcqeauxz)Live stream for Computing, Alumni and OMSCS
6:30 – 9 p.m. | Reception and Monte Carlo Night - Alumni, Students, Faculty, and Staff - Klaus Atrium and Room 1116
Title: Developing Robust Models, Algorithms, Databases, and Tools with Applications to Cybersecurity and Healthcare
Date: Wednesday, May 12, 2021
Time: 12pm-2pm EST
Location (virtual): https://bluejeans.com/8164507038/
Scott Freitas
Machine Learning Ph.D. Student
School of Computational Science and Engineering
Georgia Institute of Technology
As society and technology becomes increasingly interconnected, so does the threat landscape. Once isolated threats now pose serious concerns to highly interdependent systems, highlighting the fundamental need for robust machine learning. This dissertation contributes novel tools, algorithms, databases and models—through the lens of robust machine learning—in a research effort to solve large-scale societal problems affecting millions of people in the areas of cybersecurity and healthcare.
1. Tools: We develop TIGER, the first comprehensive graph robustness toolbox; and our Robustness Survey identifies critical yet missing areas of graph robustness research.
2. Algorithms: Our survey and toolbox reveal existing work has overlooked lateral attacks on computer authentication networks. We develop D2M, the first algorithmic framework to quantify and mitigate network vulnerability to lateral attacks by modeling lateral attack movement from a graph theoretic perspective.
3. Databases: To prevent lateral attacks altogether, we develop MalNet-Graph, the world’s largest cybersecurity graph database—containing over 1.2M graphs across 696 classes—and show the first large-scale results demonstrating the effectiveness of malware detection through a graph medium. We plan to extend MalNet-Graph by constructing the largest binary-image cybersecurity database—containing 1.2M images, 133x more images than the only other public database—enabling new discoveries in malware detection and classification research restricted to a few industry labs (MalNet-Image).
4. Models: To protect systems from adversarial attacks, we develop UnMask, the first model that flag semantic incoherence in computer vision systems, which detects up to 96.75% of attacks, and defends the model by correctly classifying up to 93% of attacks. Inspired by UnMask's ability to protect computer visions systems from adversarial attack, we develop REST, which creates noise robust models through a novel combination of adversarial training, spectral regularization and sparsity regularization. In the presence of noise, our method improves state-of-the-art sleep stage scoring by 71%--allowing us to diagnose sleep disorders earlier on and in the home environment—while using 19x less parameters and 15x less MFLOPS.
Registration is required. Register here.
In this seminar, I will talk about generative models based on point processes for financial time series simulation. Specifically, we focus on a recently developed state-dependent Hawkes (sdHawkes) process to model the limit order book dynamics [Morariu-Patrichi, 2018]. The sdHawkes model consists of an oracle Hawkes process and a state process following Markov transition. The Hawkes and state processes are fully coupled, which enables the point process captures the self-and cross-excitation as well as the interaction between events and states. We will go through the model formulation in sdHawkes, the simulation of sdHawkes, its maximum likelihood estimation, and more importantly, its application to high-frequency data modeling that captures the interactions between the order flow and the state of the current market.
Morariu-Patrichi, Maxime, and Mikko S. Pakkanen. "State-dependent Hawkes processes and their application to limit order book modelling." arXiv preprint arXiv:1809.08060 (2018).
Qi Wei received his Ph.D. degree in machine learning and image processing from the National Polytechnic Institute of Toulouse (INPENSEEIHT), University of Toulouse, France in September 2015, and Bachelor degree in Electrical Engineering from Beihang University (BUAA), Beijing, China in July 2010. Wei's doctoral thesis Bayesian Fusion of Multi-band Images: A Powerful Tool for Super-resolution was rated as one of the best theses (awarded Prix Leopold Escande) at the University of Toulouse, 2015.
Wei has worked on multiband image processing as a Research Associate with Signal Processing Laboratory, University of Cambridge, UK, and as a Research Associate at Duke University, US. He has also worked at Siemens Corporate Technology as a Research Scientist. Since 2018, Wei served as a vice president and machine learning scientist at JPMorgan. His research has been focused on machine/deep learning, time series analysis, computer vision/image processing, Bayesian statistical inference, etc.
]]>allie.mcfadden@cc.gatech.edu
]]>
A wave of recent work has sought to understand how pretrained language models work. Such analyses have resulted in two seemingly contradictory sets of results. On one hand, work based on "probing classifiers" generally suggests that SOTA language models contain rich information about linguistic structure (e.g., parts of speech, syntax, semantic roles). On the other hand, work which measures performance on linguistic "challenge sets" shows that models consistently fail to use this information when making predictions. In this talk, I will present a series of results that attempt to bridge this gap. Our recent experiments suggest that the disconnect is not due to catastrophic forgetting nor is it (entirely) explained by insufficient training data. Rather, it is best explained in terms of how "accessible" features are to the model following pretraining, where "accessibility" can be quantified using an information-theoretic interpretation of probing classifiers.
Ellie Pavlick is an Assistant Professor of Computer Science at Brown University where she leads the Language Understanding and Representation (LUNAR) Lab. She received her PhD from the one-and-only University of Pennsylvania. Her current work focuses on building more cognitively-plausible models of natural language semantics, focusing on grounded language learning and on sample efficiency and generalization of neural language models.
]]>allie.mcfadden@cc.gatech.edu
]]>
Markov decision processes (MDPs) is a minimalist framework to capture that many tasks require long-term plans and feedback due to noisy dynamics. Yet, as a result MDPs lack structure and as such planning and learning in MDPs with the typically enormous state and action spaces is strongly intractable; no algorithm can avoid Bellman's curse of dimensionality in the worst case. However, as recognized already by Bellman and his co-workers at the advent of our field, for many problem of practical interest, the optimal value function of an MDP is well approximated by just using a few basis functions, such as those that are standardly used in numerical calculations. As knowing the optimal value function is essentially equivalent to knowing how to act optimally, one hopes that this observation can be turned into efficient algorithms as there are only a few coefficients to compute. If this is possible, we can think of the resulting algorithms as performing computations with a compressed form of the value functions. While many algorithms have been proposed as early as in the 1960s, until recently not much has been known about whether these compressed computations are possible and when. In this talk, I will discuss a few recent results (some positive, some negative) that are concerned with these compressed computations and conclude with some open problems. As we shall see, still today, there are more open questions than questions that have been satisfactorily answered.
Csaba Szepesvari is a Canada CIFAR AI Chair, the team-lead for the “Foundations” team at DeepMind and a Professor of Computing Science at the University of Alberta. He earned his PhD in 1999 from Jozsef Attila University, in Szeged, Hungary. He has authored three books and over 200 peer-reviewed journal and conference papers. He serves as the action editor of the Journal of Machine Learning Research and Machine Learning, as well as on various program committees. Dr. Szepesvari's interest is artificial intelligence (AI) and, in particular, principled approaches to AI that use machine learning. He is the co-inventor of UCT, a widely successful Monte-Carlo tree search algorithm. UCT ignited much work in AI, such as DeepMind's AlphaGo which defeated the top Go professional Lee Sedol in a landmark game. This work on UCT won the 2016 test-of-time award at ECML/PKDD.
]]>allie.mcfadden@cc.gatech.edu
]]>Deep learning has changed the computing paradigm. Today, AI researchers & practitioners increasingly use deep neural networks for many applications across different modalities and areas such as NLP, Vision, Speech, Conversational and Multimodal AI. However, much of the Deep Learning revolution has been limited to the Cloud and highly specialized hardware. Recently the AI community has witnessed an increasing trend for training larger and larger neural models (e.g., GPT-3, T5, BERT) that achieve state-of-the-art results but require enormous computation, memory and energy resources on the Cloud. In order to enable AI experiences in real-time across all users and devices, ML models have to run efficiently on the Cloud and personal devices on the Edge (e.g., mobile phones, wearables, IoT) which have limited computing capabilities.
In this talk, I will introduce our work on Neural Projection computing, an efficient AI paradigm, and a family of efficient Projection Neural Network architectures that yield fast (e.g., quadratic speedup for transformer networks) and tiny models that shrink memory requirements by upto 10000x while achieving near state-of-the-art performance powering vision and NLP applications on billions of mobile devices. Widespread increase in availability of connected “smart” appliances (e.g., conversational assistants) means that there is an ever-expanding surface area for mobile intelligence and ambient devices in homes. Our approach enables efficient ML to solve complex prediction tasks for such applications both on-device and on Cloud, keeping model size, compute and power usage low while simultaneously optimizing for accuracy.
Dr. Sujith Ravi is a Director at Amazon Alexa AI where he is leading efforts to build the future of multimodal conversational AI experiences at scale. Prior to that, he was leading and managing multiple ML and NLP teams and efforts in Google AI. He founded and headed Google’s large-scale graph-based semi-supervised learning platform, deep learning platform for structured and unstructured data as well as on-device machine learning efforts for products used by billions of people in Search, Ads, Assistant, Gmail, Photos, Android, Cloud and YouTube. These technologies power conversational AI (e.g., Smart Reply), Web and Image Search; On-Device predictions in Android and Assistant; and ML platforms like Neural Structured Learning in TensorFlow, Learn2Compress as Google Cloud service, TensorFlow Lite for edge devices.
Dr. Ravi has authored over 100 scientific publications and patents in top-tier machine learning and natural language processing conferences. His work has been featured in press: Wired, Forbes, Forrester, New York Times, TechCrunch, VentureBeat, Engadget, New Scientist, among others, and also won the SIGDIAL Best Paper Award in 2019 and ACM SIGKDD Best Research Paper Award in 2014. For multiple years, he was a mentor for Google Launchpad startups. Dr. Ravi was the Co-Chair (AI and deep learning) for the 2019 National Academy of Engineering (NAE) Frontiers of Engineering symposium. He was also the Co-Chair for EMNLP 2020, ICML 2019, NAACL 2019, and NeurIPS 2018 ML workshops and regularly serves as Senior/Area Chair and PC of top-tier machine learning and natural language processing conferences like NeurIPS, ICML, ACL, NAACL, AAAI, EMNLP, COLING, KDD, and WSDM.
Website: www.sravi.org
Twitter: @ravisujith
LinkedIn: https://www.linkedin.com/in/sujithravi
]]>allie.mcfadden@cc.gatech.edu
]]>TALK TITLE
Learning Tree Models in Noise: Exact Asymptotics and Robust Algorithms
ABSTRACT
We consider the classical problem of learning tree-structured graphical models but with the twist that the observations are corrupted in independent noise. For the case in which the noise is identically distributed, we derive the exact asymptotics via the use of probabilistic tools from the theory of strong large deviations. Our results strictly improve those of Bresler and Karzand (2020) and Nikolakakis et al. (2019) and demonstrate keen agreement with experimental results for sample sizes as small as that in the hundreds. When the noise is non-identically distributed, Katiyar et al. (2020) showed that although the exact tree structure cannot be recovered, one can recover a "partial" tree structure; that is, one that belongs to the equivalence class containing the true tree. We propose Symmetrized Geometric Averaging (SGA), a statistically robust algorithm for partial tree recovery. We provide error exponent analyses and extensive numerical results on a variety of trees to show that the sample complexity of SGA is significantly better than the algorithm of Katiyar et al. (2020). SGA can be readily extended to Gaussian models and is shown via numerical experiments to be similarly superior.
https://arxiv.org/abs/2101.08917
https://arxiv.org/abs/2005.04354
This is joint work with Anshoo Tandon, Aldric J. Y. Han and Shiyao Zhu.
ABOUT VINCENT
Vincent Y. F. Tan received the B.A. and M.Eng. degrees in electrical and information sciences from Cambridge University and the Ph.D. degree in electrical engineering and computer science (EECS) from the Massachusetts Institute of Technology (MIT).
He is currently a Dean’s Chair Associate Professor with the Department of Electrical and Computer Engineering and the Department of Mathematics, National University of Singapore (NUS). His research interests include information theory, machine learning, and statistical signal processing.
He was also an IEEE Information Theory Society Distinguished Lecturer in 2018/9. He is currently serving as an Associate Editor for the IEEE Transactions on Signal Processing and an Associate Editor for Machine Learning for the IEEE Transactions on Information Theory. He is a member of the IEEE Information Theory Society Board of Governors.
allie.mcfadden@cc.gatech.edu
]]>
Time: 01/29/2021, 12.30pm - 1.30pm
Location: https://primetime.bluejeans.com/a2m/live-event/eatygusw
Abstract:
I’ll present two recent projects showing the range of domains I’d like to tackle in my work to help experts with diverse real-world research questions. First, I’ll present a model capable of filling in missing parts in ancient cuneiform tablets written thousands of years ago in now-extinct languages (Akkadian and Sumerian). Due to deterioration over time, these excavated tablets are often broken, faded, or cracked, making it hard for historians and archaeologists to read and interpret them. We show that by leveraging large-scale language models pretrained on modern texts we achieve good results in restoring missing parts in various domains and time periods, in the automatic evaluation as well as human analysis. Second, I will discuss a novel document-level representation of wet lab biochemistry protocols geared towards experiment automation and reproducibility, addressing challenges such as cross-sentence relations, long-range coreference, grounding, and implicit arguments. I’ll show examples from a manually-annotated corpus of complex lab protocols, and present graph-prediction models that form the first step towards fully executable lab protocols.
Bio:
Dr. Gabriel Stanovsky is a senior lecturer (≈ assistant professor) at the Hebrew University of Jerusalem. He did his postdoctoral research at the University of Washington and the Allen Institute for AI in Seattle, working with Prof. Luke Zettlemoyer and Prof. Noah Smith, and his PhD with Prof. Ido Dagan at Bar-Ilan University. He is interested in developing NLP models that with benefits for users in real-world applications. His work has received awards at top-tier conferences, including ACL and CoNLL, and recognition in popular journals such as Science and The New York Times.
]]>mmaddela3@gatech.edu
]]>Registration is required. Register here.
Recent progress in deep generative models such as Generative Adversarial Networks (GANs) has enabled synthesizing photo-realistic images, such as faces and scenes. However, it remains much less explored on what has been learned in the deep generative representation and why diverse realistic images can be synthesized. In this talk, I will present our recent series work from GenForce (https://genforce.github.io/) on interpreting and utilizing latent space of the GANs. Identifying these semantics not only allows us to better understand the inner working of the deep generative models but also facilitates versatile image editings. I will also briefly talk about the inverse problem (how to invert a given image into the latent code) and the fairness of the generative model.
Bolei Zhou is an Assistant Professor with the Information Engineering Department at the Chinese University of Hong Kong. He earned his PhD in computer science at the Massachusetts Institute of Technology. His research is on machine perception and autonomy, with a focus on enabling interpretable human-AI interactions. He received the MIT Tech Review’s Innovators under 35 in Asia-Pacific Award, Facebook Fellowship, Microsoft Research Asia Fellowship, MIT Greater China Fellowship, and his research was featured in media outlets such as TechCrunch, Quartz, and MIT News.
Communications Officer
allie.mcfadden@cc.gatech.edu
]]>
Date: Tuesday, January 12, 2021
Time: 1:30 pm – 3:00 pm (EST)
Location: https://bluejeans.com/5058397778
Yanbo Xu
Machine Learning PhD Student
School of Computational Science and Engineering
College of Computing
Georgia Institute of Technology
Dr. Chao Zhang (Advisor) – School of Computational Science and Engineering, Georgia Institute of Technology
Dr. Yao Xie – H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology
Dr. Yajun Mei – H. Milton Stewart School of Industrial and System Engineering, Georgia Institute of Technology
With the wide adoption of Electronic Health Record (EHR) systems across the states in the past decade, vast amount of patient data become available for machine learning researchers to develop accurate models that can potentially improve patient managements and outcomes. Developing robust and real-time computational tools for assisting critical care in the intensive care units (ICUs) is particularly of great value because of the high cost (taking about 15-20% of hospital budgets) of the care, high mortality rate (ranging from 8-19% or about 500,000 deaths annually) in the units, and the extremely large detailed data that have been collected but mostly wasted. That is, an overwhelming amount of devices such as vital sign monitors, mechanical ventilators, dialysis machines, etc, are often used in the ICUs to monitor all aspects of patients, but even the most experienced and knowledgeable intensivists hardly can understand such massive information on a continuous basis. Thus we have completed works on developing dynamic machine learning models and recurrent neural networks that can accurately detect patient deterioration and serve feedbacks in real time from integrating the large amount of multimodal continuous monitoring data.
In this proposal, we focus on developing robust detection methods that customize reliable feedbacks based on individual patients. We propose two ways to achieve the goal: model calibration and personalized thresholding. With the proposed scalable calibration method, our model can generate reliable confidence scores that can be later used for raising alarms only if the prediction confidence is high. Combining with the proposed personalized thresholding method, these generated alarms can be customized and take account of the trade-off between early detection and low false alarm rates.
]]>stephanie.niebuhr@cc.gatech.edu
]]>Registration is required. Register here.
Eightfold's mission statement is to find the right career for the right person. In order to successfully find the right career for any individual, it is imperative to have a deep understanding of their past and current accomplishments and use them to forecast their future capabilities and potential. The Machine Learning at Eightfold tackles this exact problem that helps to hire for potential and answer the question "What’s Next for You"
Sanjeet Hajarnis is a Principal Engineer at Eightfold AI and focuses on building scalable Machine Learning platforms at Eightfold. At Eightfold, he focuses on building large scale language models. Prior to Eightfold, Sanjeet has worked at Uber (Matching) and Facebook (News Feed). He graduated from Georgia Tech in 2012 with a specialization in Machine Learning.
]]>Communications Officer
allie.mcfadden@cc.gatech.edu
]]>Date: Wednesday, December 9th, 2020
Time: 13:00 to 15:00 (EDT)
BlueJeans: https://bluejeans.com/225189060
Harsh Shrivastava
Machine Learning PhD Student
School of Computational Science and Engineering
Georgia Institute of Technology
I will go over two novel and generic approaches for designing deep learning architectures which incorporate our domain knowledge about the problem under consideration. The 'Cooperative Neural Networks' take their inductive biases from the underlying probabilistic graphical models, while the problem dependent 'Unrolled Algorithms' are designed using the structure obtained by unrolling an optimization algorithm on an objective function of interest as a template. We found that the neural network architectures obtained from our approaches typically end up with very few learnable parameters and provide considerable improvement in run-time compared to other deep learning methods. We have applied our techniques to solve NLP related tasks and problems in finance, healthcare & computational biology.
There are three components of my thesis:
Firstly, I will go through the Cooperative Neural Network (CoNN-sLDA) approach which we developed for the document classification task. We use the popular Latent Dirichlet Allocation graphical model as the inductive bias for the CoNN-sLDA model. We demonstrate a 23% reduction in error on the challenging MultiSent data set compared to state-of-the-art.
Secondly, I will explain the idea of using ‘Unrolled Algorithms’ for the sparse graph recovery task. We propose a deep learning architecture, GLAD, which uses an Alternating Minimization algorithm as our model inductive bias and learns the model parameters via supervised learning. We show that GLAD learns a very compact and effective model for recovering sparse graphs from data.
Finally, I will walk through our approach of solving problems related to single-cell RNA sequencing data. Specifically, we design a novel gene regulatory network reconstruction framework called `GRNUlar'. Our method smartly utilizes the expressive ability of neural networks in a multi-task learning framework merged with our `Unrolled Algorithms' technique. To the best of our knowledge, our work is the first to introduce the successful use of expression data simulators for supervised learning of gene regulatory networks from single cell RNA seq data.
]]>Register at: https://primetime.bluejeans.com/a2m/register/kgjkcdxt
Bringing Visual Memories to Life
Photography allows us to capture and share memorable moments of our lives. However, 2D images appear flat due to the lack of depth perception and may suffer from poor imaging conditions such as taking photos through reflecting or occluding elements. In this talk, I will present our recent efforts to overcome these limitations. Specifically, I will cover our recent for creating compelling 3D photography, removing unwanted obstructions seamlessly from images or videos, and estimating consistent video depth for advanced video-based visual effects. I will conclude the talk with some ongoing research and research challenges ahead.
- Jia-Bin Huang, Assistant Professor, Virginia Tech
- Website: https://filebox.ece.vt.edu/~jbhuang/
Jia-Bin Huang is an Assistant Professor in the Bradley Electrical and Computer Engineering at Virginia Tech. He received his Ph.D. degree from the Department of Electrical and Computer Engineering at the University of Illinois, Urbana-Champaign. His research interests include computer vision, computer graphics, and machine learning with a focus on visual analysis and synthesis with physically grounded constraints. His research received the best student paper award in IAPR International Conference on Pattern Recognition (ICPR) for the work on computational modeling of visual saliency and the best paper award in the ACM Symposium on Eye Tracking Research & Applications (ETRA) for work on learning-based eye gaze tracking. Huang is the recipient of the NSF CRII award, Samsung Global Outreach Award, 3M non-tenured faculty award, and a Google faculty research award.
]]>allie.mcfadden@cc.gatech.edu
]]>Participating labs include:
Time: 3:00pm-4:00pm Eastern Time (ET)
Join virtually: https://primetime.bluejeans.com/a2m/live-event/xvrbevrh
LinkedIn Data Science - Creating Global Economic Opportunities
At LinkedIn, data plays an essential role in achieving our vision of creating economic opportunity for every member of the global workforce. In this talk, Ya Xu will share perspectives from her experience and will highlight a few interesting problems her team tackles, such as measuring network effects, addressing cannibalization bias on advertising using budget split, forecasting, data privacy, responsible design, and fairness in our products and ML models.
Ya Xu, VP of Engineering and Head of Data Science Team at LinkedIn
Ya joined LinkedIn in 2013 and has since truly helped make LinkedIn a Data First company. She leads an exceptional team of talented data scientists whose work covers metrics, insights, inference and algorithms and they tackle data science challenges across product, sales, marketing, economics, infrastructure, and operations. This centralized group has 300+ data scientists distributed across US (Sunnyvale, Mountain View, San Francisco, New York), India, China, Singapore and Dublin, Ireland. Ya is passionate about bridging science and engineering to create impactful results. She and her team try to keep LinkedIn on the cutting edge while ensuring that its A.I. systems avoid providing biased results while maintaining user privacy. They help the company take active responsibility over the data they collect to ensure fairness and protect privacy.
In addition to her work at LinkedIn, Ya’s contributions outside of her day job, such as the bookshe co-authored on Experimentation and her Stanford commencement speech, are meaningful to the entire industry as well as future Data Scientists. Before LinkedIn, she worked at Microsoft and received a PhD in Statistics from Stanford University.
Ya Xu’s LinkedIn Profile: https://www.linkedin.com/in/ya-xu/
Polo Chau
Associate Professor, CSE, College of Computing
Associate Director, MS Analytics
Director of Industry Relations, Institute for Data Engineering and Science (IDEaS)
Associate Director of Corporate Relations, Center for Machine Learning (ML@GT)
2018-2021 Provost Teaching and Learning Fellow
]]>Associate Professor, CSE, College of Computing
Associate Director, MS Analytics
Director of Industry Relations, Institute for Data Engineering and Science (IDEaS)
Associate Director of Corporate Relations, Center for Machine Learning (ML@GT)
2018-2021 Provost Teaching and Learning Fellow
]]>The panel will feature thought leaders from Google, Georgia Tech, and Queer in AI, who will together answer questions like "What implications and problems exist or will exist if the tech workforce does not become more diverse?" and "How does anyone make sure they are not introducing their bias into a given system? What questions should we be asking or actions should we be taking to avoid this?"
Registration is required and available here.
Charles Isbell is the Dean of Computing and The John P. Imlay Jr. Chair at Georgia Tech’s College of Computing. He is also a professor in the School of Interactive Computing and the Machine Learning Center at Georgia Tech (ML@GT.)
Isbell's research passion is artificial intelligence. In particular, he focuses on applying statistical machine learning to building autonomous agents that must live and interact with large numbers of other intelligent agents, some of whom may be human.
Lately, Isbell has turned his energies toward adaptive modeling, especially activity discovery (as distinct from activity recognition); scalable coordination; and development environments that support the rapid prototyping of adaptive agents. As a result, he has begun developing adaptive programming languages, worrying about issues of software engineering, and trying to understand what it means to bring machine learning tools to non-expert authors, designers, and developers.
Rapha Gontijo Lopes is a research associate at Google Brain and a founder of Queer in AI. He joined Google as an AI Resident in the summer of 2018 after completing his B.S. degree in Computer Science at Georgia Tech. His work investigates how deep learning works (or doesn't), and the role that input distributions have on model robustness. As an organizer for Queer in AI, he creates academic workshops that drive the research conversation at the intersection of AI and LGBTQ+ people.
Tiffany Deng leads the Responsible AI Program Management Team at Google where she is focused on helping people build products that work for everyone. Prior to Google, Tiffany worked as a Privacy Program Manager at Facebook and as consultant in Washington, D.C.
Deven Desai is an associate director of the Machine Learning Center (ML@GT) and associate professor in the Scheller College of Business. He was also the first, and to date, only Academic Research Counsel at Google, Inc., and a Visiting Fellow at Princeton University's Center for Information Technology Policy.
Professor Desai's scholarship examines how business interests, new technology, and economic theories shape privacy and intellectual property law and where those arguments explain productivity or where they fail to capture society's interest in the free flow of information and development
]]>Communications Officer
allie.mcfadden@cc.gatech.edu
]]>Title: Towards High Precision Text Generation
Abstract:
Despite large advances in neural text generation in terms of fluency, existing generation techniques are prone to hallucination and often produce output that is unfaithful or irrelevant to the source text. In this talk, we take a multi-faceted approach to this problem from 3 aspects: data, evaluation, and modeling.
From the data standpoint, we propose ToTTo, a tables-to-text-dataset with high quality annotator revised references that we hope can serve as a benchmark for high precision text generation task. While the dataset is challenging, existing n-gram based evaluation metrics are often insufficient to detect hallucinations. To this end, we propose BLEURT, a fully learnt end-to-end metric based on transfer learning that can quickly adapt to measure specific evaluation criteria. Finally, we propose a model based on confidence decoding to mitigate hallucinations.
Collaborators: This is joint work with Thibault Sellam, Ran Tian, Xuezhi Wang, Sebastian Gehrmann, Manaal Faruqui, Bhuwan Dhingra, Diyi Yang, and Dipanjan Das.
About the author:
Ankur Parikh is a senior research scientist at Google NYC and adjunct assistant professor at NYU. His research interests are in natural language processing and machine learning with a recent focus on high precision text generation. Ankur received his PhD from Carnegie Mellon in 2015 and has received a best paper runner up award at EMNLP 2014 and a best paper in translational bioinformatics at ISMB 2011.
]]>allie.mcfadden@cc.gatech.edu
]]>In a world of abundant information targeting multiple senses, and increasingly powerful media, we need new mechanisms to model content. Techniques for representing individual channels, such as visual data or textual data, have greatly improved, and some techniques exist to model the relationship between channels that are “mirror images” of each other and contain the same semantics. However, multimodal data in the real world contains little redundancy; the visual and textual channels complement each other. We examine the relationship between multiple channels in complex media, in two domains, advertisements and political articles.
First, we collect a large dataset of advertisements and public service announcements, covering almost forty topics (ranging from automobiles and clothing, to health and domestic violence). We pose decoding the ads as automatically answering the questions “What should do viewer do, according to the ad” (the suggested action), and “Why should the viewer do the suggested action, according to the ad” (the suggested reason). We train a variety of algorithms to choose the appropriate action-reason statement, given the ad image and potentially a slogan embedded in it. The task is challenging because of the great diversity in how different users annotate an ad, even if they draw similar conclusions. One approach mines information from external knowledge bases, but there is a plethora of information that can be retrieved yet is not relevant. We show how to automatically transform the training data in order to focus our approach’s attention to relevant facts, without relevance annotations for training. We also present an approach for learning to recognize new concepts given supervision only in the form of noisy captions.
Second, we collect a dataset of multimodal political articles containing lengthy text and a small number of images. We learn to predict the political bias of the article, as well as perform cross-modal retrieval despite large visual variability for the same topic. To infer political bias, we use generative modeling to show how the face of the same politician appears differently at each end of the political spectrum. To understand how image and text contribute to persuasion and bias, we learn to retrieve sentences for a given image, and vice versa. The task is challenging because unlike image-text in captioning, the images and text in political articles overlap in only a very abstract sense. We impose a loss requiring images that correspond to similar text to live closeby in a projection space, even if they appear very diverse purely visually. We show that our loss significantly improves performance in conjunction with a variety of existing recent losses. We also propose new weighting mechanisms to prioritize abstract image-text relationships during training.
Adriana Kovashka is an Assistant Professor in Computer Science at the University of Pittsburgh. Her research interests are in computer vision and machine learning. She has authored eighteen publications in top-tier computer vision and artificial intelligence conferences and journals (CVPR, ICCV, ECCV, NeurIPS, AAAI, ACL, TPAMI, IJCV) and ten second-tier conference publications (BMVC, ACCV, WACV). She has served as an Area Chair for CVPR in 2018-2021, NeurIPS 2020, ICLR 2021, AAAI 2021, and will serve as co-Program Chair of ICCV 2025. She has been on program committees for over twenty conferences and journals, and has co-organized seven workshops. Her research is funded by the National Science Foundation, Google, Amazon and Adobe.
]]>Communications Officer | allie.mcfadden@cc.gatech.edu
]]>The healthcare industry is rapidly evolving and this change is fueled by data. Now more than ever, data management ensures data-driven personalization for each individual patient. Learn how Anthem uses data to improve the lives of each of our patients. During this session, our team will discuss how data is ingested, processed, cleansed, managed, and visualized to improve healthcare for our communities.
Sanjay Vishwakarma, Sr. Director Engineering
Sanjay Vishwakarma is the Engineering Senior Director for Data and Analytics Operations within Anthem. In this role, Sanjay architects complex solutions for Anthem’s Data Lake including: data ingestion, standardization, curation, and consumption in both onPrem and Cloud based implementations. Prior to joining Anthem, Sanjay served in enterprise architecture roles spanning several industries including healthcare, insurance, technology, and the financial industry.
Shannon Miles, Director
Shannon Miles has over 23 years of data warehousing leadership with Anthem. Shannon has owned large-scale initiatives across all lines of business and multiple domains. In his current role, Shannon is the Director within the IT Program Integrity team. In this role, Shannon leads program integrity efforts utilizing multiple platforms including analytics based claims and provider analysis.
Polo Chau
Associate Professor, CSE, College of Computing
Associate Director, MS Analytics
Director of Industry Relations, Institute for Data Engineering and Science (IDEaS)
Associate Director of Corporate Relations, Center for Machine Learning (ML@GT)
http://www.cc.gatech.edu/~dchau/
]]>Associate Professor, CSE, College of Computing
Associate Director, MS Analytics
Director of Industry Relations, Institute for Data Engineering and Science (IDEaS)
Associate Director of Corporate Relations, Center for Machine Learning (ML@GT)
]]>The healthcare industry is rapidly evolving and this change is fueled by data. Now more than ever, data management ensures data-driven personalization for each individual patient. Learn how Anthem uses data to improve the lives of each of our patients. During this session, our team will discuss how data is ingested, processed, cleansed, managed, and visualized to improve healthcare for our communities.
Sanjay Vishwakarma, Sr. Director Engineering
Sanjay Vishwakarma is the Engineering Senior Director for Data and Analytics Operations within Anthem. In this role, Sanjay architects complex solutions for Anthem’s Data Lake including: data ingestion, standardization, curation, and consumption in both onPrem and Cloud based implementations. Prior to joining Anthem, Sanjay served in enterprise architecture roles spanning several industries including healthcare, insurance, technology, and the financial industry.
Shannon Miles, Director
Shannon Miles has over 23 years of data warehousing leadership with Anthem. Shannon has owned large-scale initiatives across all lines of business and multiple domains. In his current role, Shannon is the Director within the IT Program Integrity team. In this role, Shannon leads program integrity efforts utilizing multiple platforms including analytics based claims and provider analysis.
Polo Chau
Associate Professor, CSE, College of Computing
Associate Director, MS Analytics
Director of Industry Relations, Institute for Data Engineering and Science (IDEaS)
Associate Director of Corporate Relations, Center for Machine Learning (ML@GT)
http://www.cc.gatech.edu/~dchau/
]]>Associate Professor, CSE, College of Computing
Associate Director, MS Analytics
Director of Industry Relations, Institute for Data Engineering and Science (IDEaS)
Associate Director of Corporate Relations, Center for Machine Learning (ML@GT)
]]>Register: https://primetime.bluejeans.com/a2m/register/ppyhwquh
This event will take place on Friday, October 30 at 2 pm ET// 11 am PT.
Emerging technologies, particularly within computer vision, photogrammetry, and spatial computing, are unlocking new forms of storytelling for journalists to help people understand the world around them. In this talk, members of the R&D team at The New York Times talk about their process for researching and developing new capabilities built atop emerging research. In particular, hear how they are embracing photogrammetry and spatial computing to create new storytelling techniques that allow a reader to experience an event as close to reality as possible. Learn about the process of collecting photos, generating 3D models, editing, and technologies used to scale up to millions of readers. The team will also share their vision for these technologies and journalism, their ethical considerations along the way, and a research wishlist that would accelerate their work.
In its 169 year history, The New York Times has evolved with new technologies, publishing its first photo in 1896 with the rise of cameras, introducing the world’s first computerized news retrieval system in 1972 with the rise of the computer, and launching a website in 1996 with the rise of the internet. Since then, the pace of innovation has accelerated alongside the rise of smartphones, cellular networks, and other new technologies. The Times now has the world’s most popular daily podcast, a new weekly video series, and award-winning interactive graphics storytelling. Join us for a discussion about how our embrace of emerging technologies is helping us push the boundaries of journalism in 2020 and beyond.
Members of The New York Times R&D Team, including:
Marc Lavallee, Executive Director, R&D
Or Fleisher, Sr. Engineer
Mint Boonyapanachoti, Creative Technologist
Lana Porter, Creative Director
Mark McKeague, Technical Lead
The New York Times Research and Development team applies emerging technologies in service of our company’s mission to seek the truth and help people understand the world.
Using new technologies and formats, we develop technical capabilities for our newsroom and new forms of storytelling for our readers. As part of our method, we evaluate emerging trends in media and technology and forecast how they might play out over the next two to three years. Once we identify an opportunity, we dedicate a team to explore the space and develop products in collaboration with other parts of our organization.
We're a multidisciplinary team of journalists, creative technologists, designers, and engineers.
]]>Communications Officer
allie.mcfadden@cc.gatech.edu
]]>
Robert Nowak will give a virtual seminar on October 7, 2020. Please check back soon for registration and talk details.
Register: https://primetime.bluejeans.com/a2m/register/vzjxvxjh
The field of Machine Learning (ML) has advanced considerably in recent years, but mostly in well-defined domains using huge amounts of human-labeled training data. Machines can recognize objects in images and translate text, but they must be trained with more images and text than a person can see in nearly a lifetime. The computational complexity of training has been offset by recent technological advances, but the cost of training data is measured in terms of the human effort in labeling data. People are not getting faster nor cheaper, so generating labeled training datasets has become a major bottleneck in ML pipelines.
Active ML aims to address this issue by designing learning algorithms that automatically and adaptively select the most informative examples for labeling so that human time is not wasted labeling irrelevant, redundant, or trivial examples. This talk explores the development of active ML theory and methods over the past decade, including a new approach applicable to kernel methods and neural networks, which views the learning problem through the lens of representer theorems. This perspective highlights the effect that adding a given training example has on the representation. The new approach is shown to possess a variety of desirable mathematical properties that allow active learning algorithms to learn good classifiers from few labeled examples.
Nowak holds the Nosbusch Professorship in Engineering at the University of Wisconsin-Madison, where his research focuses on signal processing, machine learning, optimization, and statistics.
]]>Communications Officer
allie.mcfadden@cc.gatech.edu
]]>Register at: https://primetime.bluejeans.com/a2m/register/rujhjqhe
Machine Learning Challenges at Google Ads
While we have made significant advances in the last two decades, serving the most relevant ads to our users is still a big challenge to date. In particular, commercial contents are full of images, videos in addition to textual information, and understanding our advertiser products and offers from their ad creatives and landing pages poses interesting multimodal modeling problems. In this talk, we will discuss a new multimodal task (Multi-Image Summarization) and the associated dataset we are releasing. We will also talk about our recent work which co-embeds text and image in a shared embedding space to improve a cross-modal retrieval task. Then we will share some challenges & experiences from quality improvements in Search Ads products.
Pradyumna Narayana is a software engineer for Google. He joined Google in 2018 and is conducting research at the intersection of computer vision and natural language processing for Search Ads. He earned his PhD in the area of Computer Vision from Colorado State University in 2018. His Ph.D work focused on Gesture recognition from videos using Deep Learning. Prior to that, he earned his MS from Colorado State University in 2015.
Kazoo Sone is a software engineer for Google. Since he joined Google in 2011, he has led several machine learning & natural language processing projects for Search Ads and Research and contributed to many of Google’s Ads products. He earned his MS from Georgia Tech and Ph.D from Caltech.
]]>Communications Officer
allie.mcfadden@cc.gatech.edu
]]>Register: https://primetime.bluejeans.com/a2m/register/ghfqxrfr
Using rationales and influential training examples to (attempt to) explain neural predictions in NLP
Modern deep learning models for natural language processing (NLP) achieve state-of-the-art predictive performance but are notoriously opaque. I will discuss recent work looking to address this limitation. I will focus specifically on approaches to: (i) Providing snippets of text (sometimes called "rationales") that support predictions, and; (ii) Identifying examples from the training data that influenced a given model output.
Byron Wallace is an assistant professor in the Khoury College of Computer Sciences at Northeastern University. He earned his PhD from Tufts University in 2012, after which he taught at Brown University as research faculty. He joined Northeastern from the University of Texas at Austin, where he was an assistant professor in the School of Information from 2014-2016.
Wallace’s research areas include artificial intelligence, data science, machine learning, natural language processing, and information retrieval, with emphasis on applications in health informatics. Byron is a member of the applied machine learning group and the Data Science and Analytics Lab at Northeastern.
Wallace develops machine learning and natural language processing methods that make synthesizing the vast biomedical evidence-base more efficient. He also works on core machine learning and natural language processing methods, with his more of his recent work concerning Convolutional Neural Network (CNN) architectures for text. Wallace has recently been developing hybrid, interactive human/machine learning systems that aim to robustly combine human and machine intelligence.
His work has been supported by grants from the Army Research Office, the NIH, and the NSF. He won the Tufts University 2012 Outstanding Graduate Researcher award and his thesis work was recognized as The Runner Up for the 2013 ACM Special Interest Group on Knowledge Discovery and Data Mining (SIG KDD) Dissertation Award. He recently co-authored the winning submission for the Health Care Data Analytics Challenge at the 2015 IEEE International Conference on Healthcare Informatics.
]]>Communications Officer
allie.mcfadden@cc.gatech.edu
]]>The event will virtually take place via Bluejeans Events and requires registration.
Nicoleta Serban is the Virginia C. and Joseph C. Mello Professor in the H. Milton Stewart School of Industrial and Systems Engineering at Georgia Tech.
Dr. Serban's most recent research focuses on model-based data mining for functional data, spatio-temporal data with applications to industrial economics with a focus on service distribution and nonparametric statistical methods motivated by recent applications from proteomics and genomics.
She received her B.S. in Mathematics and an M.S. in Theoretical Statistics and Stochastic Processes from the University of Bucharest. She went on to earn her Ph.D. in Statistics at Carnegie Mellon University. (Nicoleta's Work)
Aditya Prakash recently joined Georgia Tech as an associate professor in the School of Computational Science and Engineering. He has published one book, more than 80 papers in major venues, holds two U.S. patents and has given four tutorials at leading conferences. His work has received a best paper award and four best-of-conference selections. Tools developed by his group have been in use in many places including ORNL, Walmart and Facebook. His research interests include Data Science, Machine Learning and AI, with emphasis on big-data problems in large real-world networks and time-series. (Aditya's Work)
Munmun DeChoudhury is an associate professor in the School of Interactive Computing. She is affiliated with ML@GT, GVU Center, and IPaT. At Georgia Tech, she leads the Social Dynamics and Wellbeing Lab to study, analyze, and appropriate social media, responsibly and ethically to derive computational, large-scale data-driven insights, and to develop mechanisms and technologies for improving our well-being, particularly our mental health. Her research has been motivated by how the availability of large-scale online social data, with the amalgamation of advances in machine learning and grounding in human-centered approaches can help us answer fundamental questions relating to our social lives. (Munmun's Work)
Srijan Kumar is an assistant professor in the School of Computational Science and Engineering. His research develops data science solutions to address the high-stakes challenges on the web and in the society. He has pioneered the development of user models and network science tools to enhance the well-being and safety of people. His research has been the subject of a documentary and has been recognized with best paper awards at WWW and ICDM. (Srijan's Work) (Srijan's Work Part 2)
Kenneth Miller is a student in Georgia Tech’s Online Master’s of Computer Science (OMSCS) program. He is a partner at Erskine Law where he represents Ford Motor Company and cases in the field of mass toxic tort litigation. Miller is a 13-year veteran of the United States Navy. (Kenneth's Work)
Irfan Essa is the executive director of ML@GT and distinguished professor and senior associate dean in the School of Interactive Computing. Essa is also a senior staff research scientist at Google. His research focuses on computer vision, machine learning, computer graphics, computational perception, robotics, computer animation, and social computing. Essa is a IEEE Fellow and has published over 200 scholarly articles with several winning best paper awards.
]]>Communications Officer
allie.mcfadden@cc.gatech.edu
]]>Talk Title
Check back soon for updated information
Abstract
Check back soon for updated information
Bio
Check back soon for updated information
]]>khanson@cc.gatech.edu
]]>It's Time to Reason
The fundamental issue underlying natural language understanding is that of semantics – there is a need to move toward understanding natural language at an appropriate level of abstraction in order to support natural language understanding and communication.
Machine Learning has become ubiquitous in our attempt to induce semantic representations of natural language and support decisions that depend on it; however, while we have made significant progress over the last few years, it has focused on classification tasks for which we have large amounts of annotated data. Supporting high-level decisions that depend on natural language understanding is still beyond our capabilities, partly since most of these tasks are very sparse and generating supervision signals for these tasks does not scale.
I will discuss some of the challenges underlying reasoning – making natural language understanding decisions that depend on multiple, interdependent, models, and exemplify it using the domain of Reasoning about Time, as it is expressed in natural language.
Dan Roth is the Eduardo D. Glandt Distinguished Professor at the Department of Computer and Information Science, University of Pennsylvania, and a Fellow of the AAAS, the ACM, AAAI, and the ACL.
In 2017 Roth was awarded the John McCarthy Award, the highest award the AI community gives to mid-career AI researchers. Roth was recognized “for major conceptual and theoretical advances in the modeling of natural language understanding, machine learning, and reasoning.”
Roth has published broadly in machine learning, natural language processing, knowledge representation and reasoning, and learning theory, and has developed advanced machine learning-based tools for natural language applications that are being used widely. Until February 2017 Roth was the Editor-in-Chief of the Journal of Artificial Intelligence Research (JAIR).
Roth is a co-founder and the chief scientist of NexLP, Inc., a startup that leverages the latest advances in Natural Language Processing (NLP), Cognitive Analytics, and Machine Learning in the legal and compliance domains.
Prof. Roth received his B.A Summa cum laude in Mathematics from the Technion, Israel, and his Ph.D. in Computer Science from Harvard University in 1995.
]]>astroup@cc.gatech.edu
]]>]]>
khanson@cc.gatech.edu
]]>Dean Isbell will share his vision for shaping the future of the college he is leading, insights he has gained from connecting with alumni around the country, and thoughts on the student community. The event has an open Q&A from the audience as well.
Lunch will be provided. RSVPs at: https://forms.gle/9psszSSdmbcKE1hi7.
The College of Computing Student Council is a student-led organization that works closely with the administration of Computing and students to establish a direct communication channel between them, enabling students to help shape the future of the college.
]]>Global Optimality Guarantees for Policy Gradient Methods
Policy gradients methods are perhaps the most widely used class of reinforcement learning algorithms. These methods apply to complex, poorly understood, control problems by performing stochastic gradient descent over a parameterized class of polices. Unfortunately, due to the multi-period nature of the objective, policy gradient algorithms face non-convex optimization problems and can get stuck in suboptimal local minima even for extremely simple problems. This talk with discus structural properties – shared by several canonical control problems – that guarantee the policy gradient objective function has no suboptimal stationary points despite being non-convex. Time permitting, I’ll also discuss (1) convergence rates that follow as a consequence of this theory and (2) consequences of this theory for policy gradient performed with highly expressive policy classes.
* This talk is based on ongoing joint work with Jalaj Bhandari.
Russo joined the Decision, Risk, and Operations division of the Columbia Business School as an assistant professor in Summer 2017. Prior to joining Columbia, he spent one great year as an assistant professor in the MEDS department at Northwestern's Kellogg School of Management and one year at Microsoft Research in New England as Postdoctoral Researcher. Russo recieved his Ph.D. from Stanford University in 2015, where he was advised by Benjamin Van Roy. In 2011 Russo recieved his BS in Mathematics and Economics from the University of Michigan.
Russo's research lies at the intersection of statistical machine learning and sequential decision-making, and contributes to the fields of online optimization, reinforcement learning, and sequential design of experiments. He is interested in the design and analysis of algorithms that learn over time to make increasingly effective decisions through interacting with a poorly understood environment.
]]>khanson@cc.gatech.edu
]]>The event is from 11:30am-2:30pm, with 3 separate 1-hour sessions. Anyone interested in general employment or careers at Google; accessible software product design; or special Google programs for inclusive employment, are especially welcome. This event will serve refreshments and is sponsored by Google Accessibility and the Georgia Tech ABLE Alliance.
Free pre-registration is kindly requested. Use below link or QR code on the attached flyer. Each session is different, so you may sign-up for one or multiple sessions: https://docs.google.com/forms/d/e/1FAIpQLSc-lf_7oKv3zdc8hlmt-OPxpCJ-1y4Hk2-ejkQXkhFdcYNarQ/viewform
]]>Assistant Professor
cassie.mitchell@bme.gatech.edu
]]>RSVP Here by Monday, February 10
Solving the Flickering Problem in Modern Convolutional Neural Networks
Deep Learning has revolutionized the AI field. Despite this, there is much progress needed to deploy deep learning in safety critical applications (such as autonomous aircraft). This is because current deep learning systems are not robust to real-world nuisances (e.g., viewpoint, illumination, partial occlusion). In this talk, we take a step in constructing robust deep learning systems by addressing the problem that state-of-the-art Convolution Neural Networks (CNN) classifiers and detectors are vulnerable to small perturbations, including shifts of the image or camera. While various forms of specially engineered “adversarial perturbations” that fool deep learning systems have been well documented, modern CNNs can surprisingly change classification up to 30% probability even for simple 1-pixel shifts of the image. This lack of translational stability seems to be partially the cause of “flickering” in state-of-the-art object detectors applied to video. In this talk, we introduce this phenomena, propose a solution, prove it analytically, validate it empirically, and explain why existing CNNs exhibit this phenomena.
Ganesh Sundaramoorthi is currently Principal Research Scientist at United Technologies Research Center in East Hartford, CT, USA, conducting research in computer vision and machine learning, and building products in robotic inspection from this research. Prior to this, he was Assistant Professor of Electrical Engineering and jointly Assistant Professor of Applied Mathematics and Computational Science at KAUST starting in 2011. He directed the Computational Vision Lab at KAUST, which developed novel mathematics and algorithms, as well as software for video and image understanding technology. His fundamental optimization algorithms have led to advancements in motion-based video segmentation and detection. His group also developed technology for seismic image analysis, electron microscopy images, and medical (MRI & CT) images.
Prior to KAUST, he was a postdoctoral research associate with Prof. Stefano Soatto in the Vision Lab at the University of California, Los Angeles from 2008 to 2010. There he made fundamental contributions to the view invariance problem in object recognition (with Prof. Peter Petersen and Prof. V. S. Varadarajan), and developed technology for video tracking. His PhD is in Electrical and Computer Engineering from the Georgia Institute of Technology in Atlanta, GA, USA in 2008. His PhD developed fundamental shape optimization methods for computer vision that aided in technology for video tracking, and medical image analysis. He was advised by Prof. Anthony Yezzi. His Bachelor's degrees were in Computer Engineering and Applied Mathematics, which he earned in 2003, also from Georgia Tech.
He has served as Area Chair for the leading computer vision conferences, including IEEE Conference on Computer Vision and Pattern Recognition (CVPR) and IEEE International Conference on Computer Vision (ICCV).
]]>khanson@cc.gatech.edu
]]>Communication-Efficient Distributed Stochastic Optimization with Variance Reduction and Gradient Tracking
There is an increasing need to perform large-scale machine learning and optimization over distributed networks, e.g. in the context of multi-agent learning and federated optimization. It is well recognized that, a careful balance of local computation and global communication is necessary to fully unleash the benefits in the distributed setting. In this talk, we first consider a natural framework for distributing popular stochastic variance reduced methods in the master/slave setting, and establish its convergence guarantees under simple and intuitive assumptions that capture the effect of local data heterogeneity. Next, we move to the decentralized network setting, where each agent only aggregates information from its neighbors over a network topology. We discuss challenges and solutions to obtain decentralized counterparts for algorithms originally developed for the master/slave setting, and highlight the resulting algorithms using approximate Newton and stochastic variance-reduced local updates. Theoretical convergence guarantees and numerical evidence are provided to demonstrate the appealing performance of our algorithms over competitive baselines, in terms of both communication and computation efficiency.
Dr. Yuejie Chi received the Ph.D. degree in Electrical Engineering from Princeton University in 2012, and the B.E. (Hon.) degree in Electrical Engineering from Tsinghua University, Beijing, China, in 2007. Since 2018, she is Robert E. Doherty Career Development Professor and Associate Professor with the department of Electrical and Computer Engineering at Carnegie Mellon University, after spending 5 years at The Ohio State University. She is interested in the mathematics of data science that take advantage of structures and geometry to minimize complexity and improve performance in decision making. Specific topics include mathematical and statistical signal processing, machine learning, large-scale optimization, sampling theory, with applications in sensing, imaging and data science. She is a recipient of the PECASE Award, NSF CAREER Award, AFOSR YIP Award, ONR YIP Award, IEEE SPS Early Career Technical Achievement Award, and IEEE SPS Young Author Paper Award.
]]>khanson@cc.gatech.edu
]]>Abstract: Accurate description of protein structure and function is a fundamental step towards understanding biological life and highly relevant in the development of therapeutics. Although greatly improved, experimental protein structure determination is still low-throughput and costly, especially for membrane proteins. As such, computational structure prediction is often resorted. Predicting the structure of a protein without similar experimental structures is very challenging and usually needs a large amount of computing power. This talk will present the deep learning method (i.e., deep convolutional residual neural network) we have invented for protein contact and distance prediction that won the CASP (Critical Assessment of Structure Prediction) in both 2016 and 2018 in the category of contact prediction. In this talk we show that by using this powerful deep learning technique, even with only a personal computer we can predict the structure of a protein much more accurately than ever before. In particular, we predicted correct folds for the 3 largest hard targets (~350 amino acids) in CASP13 (2018) and generated the best 3D models for two of them among all the human and server groups including DeepMind's AlphaFold. Inspired by our success in CASP12 in 2016, this deep learning technique has been adopted widely by the structure prediction community and thus, resulted in the widespread, largest progress in the history of CASP, which will also be discussed in this talk.
Bio: Dr. Jinbo Xu is a full professor at the Toyota Technological Institute at Chicago, a computer science research and educational institute located at the University of Chicago. Dr. Xu’s research lies in machine learning, optimization, and computational biology. He has developed several popular bioinformatics programs such as the CASP-winning RaptorX (http://raptorx.uchicago.edu) for protein structure prediction and IsoRank/HubRank for interaction network analysis. The deep learning method initiated by him for protein contact/distance prediction has been widely adopted by the community and resulted in the largest progress in the history of protein structure prediction, due to which he was invited to give a keynote talk at the 2019 3DSIG session of ISMB, the largest bioinformatics conference in the world. Dr. Xu is an Associate Editor of the Bioinformatics journal and has received many awards, including Alfred P. Sloan Research Fellowship, NSF CAREER award, RECOMB's test-of-time award (2019), RECOMB best paper award (2014) and PLoS Computational Biology Research Prize (2018). His work has also been reported by Science, The Economist, and other media.
]]>astroup@cc.gatech.edu
]]>We'll walk you through the do's and don'ts of social media, how to create an amazing website, tips for becoming an all-star academic blogger, and more so that your personality and your accomplishments shine bright to recruiters, journalists, and colleagues.
Communications professionals will also be available for one-on-one feedback on your existing profiles and you can get a professional headshot!
Bring your online profiles and get ready to make your mark as you enter the job market.
P.S. there will be food.
Communications Officer
allie.mcfadden@cc.gatech.edu
]]>TITLE
NLP Approaches to Campaign Classification
ABSTRACT
Mailchimp is the world's largest marketing automation platform. Over a billion emails are sent by it every day, which raises the question: what exactly are users sending? We'll do a deep dive into the natural language processing techniques utilized by Mailchimp to make sense of users' content and classify campaigns, whether it's for predicting a customer's business vertical, or for preventing those with malicious intent from using the platform.
BIO
Muhammed Ahmed is a Junior Data Scientist at Mailchimp. He graduated from the University of Georgia with a degree in Computer Science and a certificate in Applied Data Science. Muhammed specializes in natural language processing and has implemented several deep transformer models (ELMo, BERT, XLNet, RoBERTa) for the text classification task. His models are used for spam detection, business vertical classification, and support ticket classification.
polo@gatech.edu
]]>is proud to sponsor a
Come chat with several Ralativity engineers about internships and career opportunities..
Program Manager
khanson@cc.gatech.edu
]]>is proud to present
a talk by
Director of the Vatican Observatory, Rome
About this talk
Why did we go to the Moon? Why does the Vatican support an astronomical observatory? These questions mask a deeper question: why do individuals choose to spend their lives in pursuit of pure knowledge? The motivation behind our choices, both as individuals and as a society, controls the sorts of science that gets done. It determines the kinds of answers that are found to be satisfying. And ultimately, it affects the way in which we think of ourselves.
About the speaker
Guy Consolmagno, SJ is a brother in the Roman Catholic Society of Jesus (the Jesuits), working since 1993 as an astronomer and meteorite specialist at the Specola Vaticana (Vatican Observatory), located in the Papal summer gardens outside Rome. Since 2014 he has been president of the Vatican Observatory Foundation, which supports the work of the Observatory and especially its 1.8 meter Vatican Advanced Technology Telescope (VATT) in Arizona. In September of 2015 he was named Director of the Vatican Observatory by Pope Francis.
Consolmagno's research explores connections between meteorites, asteroids, and the evolution of small solar system bodies. Along with more than 200 scientific publications, he is the author of a number of popular books, including: Turn Left at Orion (with Dan Davis), and most recently, Would You Baptize an Extraterrestrial? (with Fr. Paul Mueller, S.J.). He also has hosted science programs for BBC Radio 4, has been interviewed in numerous documentary films, and writes a monthly science column for the British Catholic magazine, The Tablet.
A native of Detroit, MI, Consolmagno earned two degrees from MIT and a doctorate in planetary sciences from the University of Arizona, was a postdoctoral research fellow at Harvard and MIT, served in the US Peace Corps (Kenya), and taught university physics at Lafayette College before entering the Jesuits in 1989. He has served as chair of the American Astronomical Society’s Division for Planetary Sciences (AAS/DPS) and on the planetary surfaces nomenclature committee of the International Astronomical Union (IAU). Asteroid “4597 Consolmagno” was named in recognition of his work. In 2014 he won the Carl Sagan Med al for public outreach by the AAS/DPS.
]]>a talk by
Friday, September 2 at 11 a.m.
SpaceX's reusable rocket program aims to reduce the cost of space travel by making rockets that can land, refuel and refly, instead of being thrown away after every flight. Precise landing of a rocket is a unique problem, which has been likened to balancing a rubber broomstick on your hand in a windstorm. Rockets do not have wings (unlike airplanes) and they cannot rely on a high ballistic coefficient to fly in a straight line (unlike missiles). In the past year, SpaceX has successfully landed five rockets, two of which were on dry land at Cape Canaveral, and three of which were on a floating platform in the Atlantic. This talk will discuss the challenges involved, how these challenges were overcome, and next steps towards rapid reusability.
]]>
Nobel Laureate Lecture: Michael Levitt
The development of multiscale models for complex chemical systems began in 1967 with publications by Warshel and Levitt, recently recognized by the 2013 Nobel Committee for Chemistry. The simplifications used then at the dawn of the age of computational structural biology were mandated by computers that were almost a billion times less cost-effective than those we use today. These same multiscale models have become increasingly popular in applications that range from simulation of atomic protein motion, to protein folding and explanation of enzyme catalysis.
In this talk, Michael Levitt will describe the origins of computational structural biology and show some of the most exciting current and future applications
Born in South Africa 1947, Michael Levitt visited London at the age of 16 to be profoundly influenced by John Kendrew's 1944 BBC TV series "The Thread of Life". After receiving a BSc in Physics at King’s College London and spending a year with Prof. Shneior Lifson and his PhD student Arieh Warshel at the Weizmann Institute in Israel, Levitt joined the Laboratory of Molecular (LMB), Cambridge, in 1968 . His PhD thesis on Protein Conformation Analysis described use of classical force-fields and introduced energy refinement.
Levitt went back to Israel as an EMBO postdoc with Lifson. His then collaboration with Warshel resulted in new multi-scale approaches to molecular modeling: coarse-grained models that merge atoms to allow folding simulation and hybrid models that combine classical and quantum mechanics to explain how enzymes works by electrostatic strain. In 1974, Levitt returned to LMB for three years, spent two years with Francis Crick at Salk and seven years at Weizmann, before moving to Structural Biology at Stanford from 1987.
His diverse interests have included RNA & DNA modeling, protein folding simulation, classification of protein folds & protein geometry, antibody modeling, x-ray refinement, antibody humanization, side-chain geometry, torsional normal mode, molecular dynamics in solution, secondary structure prediction, aromatic hydrogen bonds, structure databases, and mass spectrometry. Levitt’s current postdocs work on protein evolution, the crystallographic phase problem and Cryo-EM refinement.
While enjoying membership of the Royal Society and the National Academy, Levitt remains an active programmer, a craft skill of which he is particularly proud. His post-prize ambitions are two fold and likely inconsistent: (1) Work single-mindedly as he did in the mid-1970’s on hard problems and (2) help today’s young scientists gain the recognition and independence that Levitt’s generation enjoyed.
Married in 1968 to Rina, an active artist, they have three children and a rapidly increasing number of grandchildren, all of whom help Levitt stay more-or-less normal.
]]>Title:
Understanding the effect of social media:
Abstract:
With the emergence of social media and Web 2.0, broadcasting in the online environment has evolved into a new form of marketing due to the much broader reach enabled by information technology. This paper quantifies the effect of artists’ different broadcasting activities on a leading social media site for music, MySpace, on music sales. We employ a panel vector auto-regression (PVAR) model to investigate the inter-relationship between broadcasting promotions in social media and music sales, while controlling for influential factors such as album prices, advertising in traditional media channels, artist popularity, and the impact of user-generated content. We characterize two types of broadcast messages under the MySpace context, personal and automated. We find that broadcasting in social media has a significant effect on sales even after controlling for the aforementioned factors, and more importantly the effect mainly comes from personal messages rather than automated messages. Our findings also point to the importance of conducting captivating conversations with customers in the organizational use of social media.
Bio:
Yu (Jeffrey) Hu is an associate professor at Scheller College of Business, Georgia Institute of Technology. He received his Ph.D. from MIT’s Sloan School of Management. His research studies electronic commerce, mobile commerce, Internet retailing, social media, and online advertising. He has consulted for many retailing and publishing companies and European Commission. His research has been published in top journals such as Management Science, Information Systems Research, and Sloan Management Review, and has been discussed extensively and cited by many media outlets such as New York Times and National Public Radio. He is an associate editor for Management Science and Information Systems Research.
]]>
Title:
Galerkin Methods for Monotone Linear Complementarity Problems
Abstract:
Linear complementarity problems are one way to encode the first-order optimality conditions for inequality-constrained optimization or saddle-point problems. So, LCPs arise in a wide variety of areas, including machine learning, planning, game theory, and physical simulation. In all of these areas, to handle large-scale LCP instances, we need fast approximate solution methods. One promising idea is Galerkin approximation, in which we search for the best answer within the span of a given set of basis functions. For equality-constrained problems, Galerkin methods have had a long history of practical and theoretical successes. Unfortunately, the most straightforward way to apply Galerkin approximation to LCPs leads to difficulties in practice, including worse approximation errors than might be expected based on the ability of the basis to represent the desired solution. So, in this talk, I'll present a new Galerkin method for LCPs that attempts to address the above difficulties.
Bio:
Dr. Gordon is an Associate Research Professor in the Department of Machine Learning at Carnegie Mellon University, and co-director of the Department's Ph. D. program. He works on multi-robot systems, statistical machine learning, game theory, and planning in probabilistic, adversarial, and general-sum domains. His previous appointments include Visiting Professor at the Stanford Computer Science Department and Principal Scientist at Burning Glass Technologies in San Diego. Dr. Gordon received his B.A. in Computer Science from Cornell University in 1991, and his Ph.D. in Computer Science from Carnegie Mellon University in 1999.
]]>
Speaker: Dr. Yousef Saad, College of Science & Engineering Distinguished Professor in the Department of Computer Science & Engineering at the University of Minnesota
Title:
Multilevel Algebraic Preconditioning Techniques with Applications
Abstract:
Solving linear systems of equations with iterative methods is becoming more difficult due to a number of new challenges. Matrices of these systems are becoming larger, more ill-conditioned, and are often poorly structured, and indefinite. Multilevel methods have been advocated for handling some of these challenges. This talk will introduce a variety of multilevel preconditioners for solving linear systems of equations, with an emphasis on indefinite systems. We begin with the Algebraic Recursive Multilevel Solver (ARMS) and see how a class of "coarsening" schemes can be adapted to this framework.
ILU-type preconditioners have difficulties for some types of indefinite problems. We will show how they can be adapted for problems arising from Helmholtz equations. Then a new class of methods based on low-rank approximations which has some appealing features will be introduced. The methods handle indefiniteness quite well and are more amenable to SIMD computations, which makes them attractive for GPUs. We will then present an application in dynamic mean field theory (DMFT) where the problem is to compute the diagonal of the inverse of a matrix.
Bio:
Yousef Saad is a College of Science & Engineering Distinguished Professor in the Department of Computer Science & Engineering at the University of Minnesota. He holds the William Norris Chair for Large-Scale Computing and is a fellow of SIAM and AAAS. He is known for his contributions to matrix computations, including iterative methods for solving large sparse linear algebraic systems, eigenvalue problems, and parallel computing. Dr. Saad is an ISI highly cited researcher in mathematics and is the author of the highly cited book, Iterative Methods for Sparse Linear Systems. For more information, please visit http://www-users.cs.umn.edu/~saad
]]>
Speaker: Sharon Aviran, Center for Computational Biology at UC Berkeley
Date: Tuesday, February 26, 2013
Location: Klaus 1456
Time: 11:00am-12:00pm
-----------------------------------
Title:
Modeling and High-Throughput Analysis of RNA Structure Mapping Experiments
Abstract:
New regulatory roles continue to emerge for both natural and engineered RNAs, many of which have specific structures essential to their function. This highlights a growing need to develop technologies that enable rapid and accurate characterization of RNA structure. Yet, available techniques that are reliable are also vastly limited, while the accuracy of popular computational methods is generally poor. These limitations thus pose a major barrier to comprehensive determination of structure from sequence.
To address this need, we have developed a high-throughput structure characterization assay, called SHAPE-Seq, which simultaneously measures structural information at nucleotide-resolution for hundreds of distinct RNAs. SHAPE-Seq combines a novel chemistry with next-generation sequencing of its products. Following sequencing, we extract the structural information using a fully automated algorithmic pipeline that we developed. In this talk, I will focus on SHAPE-Seq's analysis methodology, which relies on a novel probabilistic model of a SHAPE-Seq experiment, adjoined by maximum-likelihood parameter estimation. I will demonstrate the accuracy, simplicity, and efficiency of our approach, and will then present an algorithm that uses such structure mapping data to inform computational RNA secondary structure prediction.
Bio
Sharon Aviran is a postdoctoral researcher at the Center for Computational Biology at UC Berkeley, working with Prof. Lior Pachter. Her current research interests are in the areas of Genomics and Functional Genomics, focusing on developing computational methods for high-throughput analysis of RNA molecular dynamics. She pursued her PhD in Electrical Engineering at UCSD, working with Professors Paul Siegel and Jack Wolf, and specialized in signal processing for communications and in information theory. She was awarded the 2006 Sheldon Schultz Prize for Excellence in Graduate Research, and was a Calit2 Fellow at UCSD and a Postdoctoral Innovation Fellow at UC Berkeley. In 2012, she received the NIH K99/R00 career development award for her research on RNA structural dynamics.
-----------------------------------
]]>
Title:
The Science of Ice Sheets: the Mathematical Modeling and Computational Simulation of Ice Flows
Abstract:
The melting of ice in Greenland and Antarctica would, of course, be by far the major contributor to sea level rise. Thus, to make science-based predictions about sea-level rise, it is crucial that the ice sheets covering those land masses be accurately mathematically modeled and computationally simulated. In fact, the 2007 IPCC report on the state of the climate did not include predictions about sea level rise because it was concluded there that the science of ice sheets was not developed to a sufficient degree. As a result, such predictions could not be rationally and confidently made. In recent years, there has been much activity in trying to improve the state-of-the-art of ice sheet modeling and simulation. In this lecture, we review a hierarchy of mathematical models for the flow of ice, pointing out the relative merits and demerits of each, showing how they are coupled to other climate system components, and discussing where further modeling work is needed. We then discuss algorithmic approaches for the approximate solution of ice-sheet flow models and present and compare results obtained from simulations using the different mathematical models.
Bio:
Max Gunzburger is the Frances Eppes Eminent Professor and Founding Chair of the Department of Scientific Computing at Florida State University. He has received numerous awards and honors including the W. T. and Idelia Reid Prize in Mathematics from the Society for Industrial and Applied Mathematics (SIAM), being a charter fellow of SIAM, and a NASA Innovator's Prize for Inventions and Contributions. He received the Rostchild Visiting Fellow award from Cambridge University and the OCCAM Visiting Fellow award from Oxford University and serves as a CSRI Senior Research Fellow at the Sandia National Laboratories.
Max Gunzburger received his B.S., M.S., and Ph.D. from New York University. He previously served as a Distinguished Professor of Mathematics and Chair of the Mathematics Department at Iowa State University and as Professor of Mathematics at Virginia Tech, Carnegie Mellon University, and the University of Tennessee. He holds a Distinguished Professor Appointment at Yonsei University in South Korea and previously served as a Guest Professor at Peking University. He has served on the editorial board numerous journals and book series, including three SIAM journals, and was Editor in Chief of the SIAM Journal on Numerical Analysis and is a Founding Editor and Senior Editor of the SIAM/ASA Journal on Uncertainty Quantification. He has served on numerous SIAM committees and was Chairman of the Board of Trustees of SIAM. He has served as a consultant to three DOE and two NASA laboratories as well as several industrial and commercial organizations.
Max Gunzburger's research interests spans the areas of numerical analysis, scientific computing, optimization and control, computational geometry, and partial differential equations with applications in diverse areas including fluid and solid mechanics, climate, materials, subsurface flows, image processing, diffusion processes, superconductivity, acoustics, electromagnetics, etc.
]]>
Speaker: Peter Volkov, Malware Analyst, Yandex, Moscow, Russia
PIZZA will be provided
-----------------------------------
Title:
Evolution of mass DbD-attacks malware and co-evolution of antimalware technologies
Abstract:
Exploitation of web browser environment vulnerabilities is one of the major ways of distrubution of mass malware.
I will talk about current trends in evolution of drive-by download related malware, juxtaposed with co-evolution in development of modern antimalware technologies to counteract DbD attacks. I'll also talk about specific exploit-kits existing in the black market as well as about the current state of antimalware industry.
Examples of actual malware codes, vulnerable features of browser environment and internals of antimalware systems will be presented and discussed.
Bio:
Peter Volkov is a Malware Analyst at the Yandex company http://company.yandex.com/ He graduated from the Faculty of Computational Mathematics and Cybernetics, Moscow State University.
Before joining Yandex Peter worked at Kaspersky's Lab on analysis of malicious codes, applied forensics and training programs for students.
]]>
Title:
Cyber Games
Abstract:
Over the last few years I have been working on game theoretic models of security, with a particular emphasis on issues salient in cyber security. In this talk I will give an overview of some of this work. I will first spend some time motivating game theoretic treatment of problems relating to cyber and describe some important modeling considerations. In the remainder, I will describe two game theoretic models (one very briefly), and associated solution techniques and analyses. The first is the "optimal attack plan interdiction" problem. In this model, we view a threat formally as a sophisticated planning agent, aiming to achieve a set of goals given some specific initial capabilities and considering a space of possible "attack actions/vectors" that may (or may not) be used towards the desired ends. The defender's goal in this setting is to "interdict" a select subset of attack vectors by optimally choosing among miti-gation options, in order to prevent the attacker from being able to achieve its goals. I will describe the formal model, explain why it is challenging, and present highly scalable decomposition-based integer programming techniques that leverage extensive research into heuristic formal planning in AI. The second model addresses the problem that defense decisions are typically decentralized. I describe a model to study the impact of decentralization, and show that there is a "sweet spot": for an intermediate number of decision makers, the joint decision is nearly socially optimal, and has the additional benefit of being robust to the changes in the environment.
Finally, I will describe the Secure Design Competition (FIREAXE) that involved two teams of interns during the summer of 2012. The problem that the teams were tasked with was to design a highly stylized version of an electronic voting system. The catch was that after the design phase, each team would attempt to "attack" the other's design. I will describe some salient aspects of the specification, as well as the outcome of this competition and lessons that we (the designers and the students) learned in the process.
Bio:
Yevgeniy Vorobeychik is a Principal Member of Technical Staff at Sandia National Laboratories. Between 2008 and 2010 he was a post-doctoral research associate at the University of Pennsylvania Computer and Information Science department. He received Ph.D. (2008) and M.S.E. (2004) degrees in Computer Science and Engineering from the University of Michigan, and a B.S. degree in Computer Engineering from Northwestern University. His work focuses on game theoretic modeling of security, algorithmic and behavioral game theory and incentive design, optimization, complex systems, epidemic control, network economics, and machine learning. Dr. Vorobeychik has published over 50 research articles on these topics, including publications in top Computer Science, Operations Research, Business, and Physics journals and conferences. Dr. Vorobeychik was nominated for the 2008 ACM Doctoral Dissertation Award and received honorable mention for the 2008 IFAAMAS Distinguished Dissertation Award. In 2012 he was nominated for the Sandia Employee Recognition Award for Technical Excellence. He was also a recipient of a NSF IGERT interdisciplinary research fellowship at the University of Michigan, as well as a distinguished Computer Engineering undergraduate award at Northwestern University.
]]>
Speaker: Prof. Chris Rycroft, UC Berkeley (Math)
-----------------------------------
Title:
Modeling the toughness of metallic glasses
Abstract:
Metallic glasses are a new type of alloy whose atoms have an amorphous arrangement in contrast to most metals. They have many favorable properties such as excellent wear resistance and high tensile strength, but are prone to breakage in some circumstances, depending on their method of preparation. The talk will describe the development of a quasi-static projection method within an Eulerian finite-difference framework, for simulating a new physical model of a metallic glass. The simulations are capable of resolving the multiple timescales that are involved, and provide an explanation of the experimentally observed differences in breakage strength, which may aid in the use of these materials in practical applications. The same Eulerian simulation framework can be adapted to address a variety of other problems, such as fluid-structure interaction, and the mechanical modeling of multicellular clusters.
Biography:
Chris Rycroft is a Morrey Assistant Professor in the Departments of Mathematics at the University of California, Berkeley and the Lawrence Berkeley National Laboratory. He is interested in mathematical modeling and scientific computation, particularly for interdisciplinary applications in science and engineering. He has recently been working on several projects relating to materials science, and since 2010 he has also been involved in the Physical Sciences in Oncology program, a pilot initiative to encourage multi-disciplinary cancer research. He obtained his PhD in Mathematics in 2007 from Massachusetts Institute of Technology under the supervision of Martin Z. Bazant.
Website: http://math.berkeley.edu/~chr/
]]>
Speaker: Dr. Jack Poulson, Postdoctoral Fellow in the Department of Mathematics at Stanford University
Title:
Fast parallel solution of heterogeneous 3D time-harmonic wave equations
Abstract:
Several advancements related to the iterative solution of heterogeneous 3D time-harmonic wave equations are presented. In particular, efficient distributed-memory parallelizations of sweeping preconditioners are discussed in the context of Helmholtz equations and
linear elasticity, and it will be shown that challenging 3D problems approaching a billion degrees of freedom can be solved in just a few minutes using several thousand cores. In addition, several high-performance parallel algorithms are proposed for performing multifrontal triangular solves with many right-hand sides, and a custom Kronecker-product compression scheme for the sweeping preconditioner is introduced which is motivated by the translation invariance of free-space Green's functions. Lastly, the software developed as part of this
research is made freely available through Parallel Sweeping Preconditioner (PSP) [1] and Clique [2], a distributed-memory multifrontal solver. Both of these packages are built upon the author's distributed-memory dense linear algebra library, Elemental [3].
Author Bio:
Jack Poulson is a Postdoctoral Fellow in the Department of Mathematics at Stanford University working with Lexing Ying. He received his Ph.D. in Computational and Applied Mathematics from the Institute of Computational Engineering and Sciences (ICES)
at the University of Texas at Austin at the end of 2012 and is interested in the development and application of scalable fast algorithms.
[1] http://bitbucket.org/poulson/psp
[2] http://bitbucket.org/poulson/clique
[3] http://code.google.com/p/elemental
]]>Speaker: Josh Patterson, Cloudera
Title:
Knitting Boar
Abstract:
Online learning techniques, such as Stochastic Gradient Descent (SGD), are powerful when applied to risk minimization and convex games on large problems. However, their sequential design prevents them from taking advantage of newer distributed frameworks such as Hadoop/YARN. In this session, we will introduce “Knitting Boar”, an open-source Java library for performing distributed online learning on a Hadoop cluster under YARN. We will give an overview of how Knitting Boar works and examine the lessons learned from YARN application construction.
Bio:
Josh Patterson is a Principal Solution Architect at Cloudera. Prior to joining Cloudera, he was responsible for bringing Hadoop into the smartgrid during his involvement in the openPDC project. His focus in the smartgrid realm with Hadoop and HBase was using machine learning to discover and index anomalies in time series data. Josh is a graduate of the University of Tennessee at Chattanooga with a Bachelors in Business Management and a Masters of Computer Science with a thesis titled "TinyTermite: A Secure Routing Algorithm" where he worked in mesh networks and social insect swarm algorithms. Josh has over 15 years in software development and continues to contribute to projects such as Apache Mahout, openPDC, and JMotif in the open source community. Currently Josh focuses on open source parallel linear modeling and optimization techniques.
]]>
Polo Chau <polo@gatech.edu>
]]>Title:
Dimensionality Reduction From Several Angles
Abstract:
I will present five past and current projects that attack the problem of dimensionality reduction (DR) from quite different methodological angles. Two projects nicely fit into the usual mold of technique-driven work on algorithms for DR. Glimmer is a multilevel multidimensional scaling (MDS) algorithm that exploits the GPU. Glint is a new MDS framework that achieves high performance on costly distance functions. In contrast, the DimStiller project is a foray into systems rather than algorithms, built around the idea of "DR for the rest of us". It is a toolkit for DR that provides local and global guidance to users who may not be experts in the mathematics of high-dimensional data analysis. A third kind of project combines evaluation and the creation of taxonomies. Our recent taxonomy of visual cluster separation factors arose from the systematic qualitative examination of over 800 scatterplots of dimensionally reduced data, and includes an analysis of the reasons for failure of previous cluster separation metrics. I will also discuss the current work of a task taxonomy that is grounded in a two-year qualitative study of high-dimensional data analysts in many domains, to discover how the use of DR "in the wild" does and does not match up with the assumptions that underlie previous algorithmic work.
Bio:
Tamara Munzner is a professor at the University of British Columbia Department of Computer Science, where she has been since 2002. She was a research scientist from 2000 to 2002 at the Compaq Systems Research Center in California, earned her PhD from Stanford between 1995 and 2000, and was a technical staff member at the Geometry Center mathematical visualization research group from 1991 to 1995. Tamara was InfoVis Co-Chair in 2003 and 2004 and EuroVis Co-Chair in 2009 and 2010. Her research interests include the development, evaluation, and characterization of information visualization systems and techniques from both user-driven and technique-driven perspectives. She has worked on visualization projects in a broad range of application domains, including evolutionary biology, microbiology, topology, computational linguistics, large-scale system administration, web site design, and web log analysis.
Webcast link:
http://proed.pe.gatech.edu/gtpe/pelive/fodava_120712/
]]>
Polo Chau, Assistant Professor, Georgia Tech CSE, College of Computing
]]>Title:
High-performance-computing challenges for heart simulations
Abstract:
The heart is an electro-mechanical system in which, under normal conditions, electrical waves propagate in a coordinated manner to initiate an efficient contraction. In pathologic states, propagation can destabilize and exhibit chaotic dynamics mostly produced by single or multiple rapidly rotating spiral/scroll waves that generate complex spatiotemporal patterns of activation that inhibit contraction and can be lethal if untreated. Despite much study, little is known about the actual mechanisms that initiate, perpetuate, and terminate spiral waves in cardiac tissue.
In this talk, I will motivate the problem with some experimental examples and then discuss how we study the problem from a computational point of view, from the numerical models derived to represent the dynamics of single cells to the coupling of millions of cells to represent the three-dimensional structure of a working heart. Some of the major difficulties of computer simulations for these kinds of systems include: i) Different orders of magnitude in time scales, from milliseconds to seconds; ii) millions of degrees of freedom over millions of integration steps within irregular domains; and iii) the need for near-real-time simulations. Advances in these areas will be discussed as well as the use of GPUs over the web using webGL.
Bio: Flavio Fenton is an associate professor in the School of Physics at Georgia Tech. He is an accomplished scholar in the area of biophysics of the heart. He received his PhD from Northeastern University. He served as director of Electrophysiology Research at The Heart Institute at Beth Israel Medical Center in NY, and also worked as a research associate in Biomedical Sciences at Cornell University just prior to joining Georgia Tech.
]]>
IDH Distinguished Lecture Speaker
By: Dan Reed Corporate Vice President, Technology Policy Group, Microsoft Corporation
Title:
HPC 2.0: The Challenge of Scale
Abstract:
In just a few years, scientific computing has moved from the terascale to the trans-petascale regime, with research on exascale designs now underway. Large-scale scientific instruments, ubiquitous sensors and large scale simulations are now producing a torrent of digital data, with associated challenges of transport, analysis and curation. Disciplinary fusion and multidisciplinary collaborations are challenging our social structures, as teams from diverse backgrounds and organizations work together. Changes in scale, whether technical or sociological, always bring new challenges.
What are the emerging technologies that will drive the future of high-performance computing? How can we best integrate software management of component failures and system resilience, energy consumption and efficiency, and performance optimization in manageable and intuitive ways? How can we measure and optimize the performance of systems at scale? How do we balance use of commodity components and economies of scale against custom design, recognizing that technical computing has always tracked the consumer mainstream? What are the best ways to manage and extract insights from data at scale, and what are the economically viable ways to enable multidisciplinary data fusion and technology transfer? How can we simplify use of technical computing and broaden the base of users?
This talk will examine these and other issues technically, socially and economically, with an eye to the fusion of technology and policy.
Bio:
I head Microsoft's Technology Policy Group, whose role is to help shape Microsoft's long-term vision for technology innovations and the company's associated policy engagement with governments and institutions around the world.
Previously, I was the founding director of the Renaissiance Computing Institute (RENCI) at the University of North Carolina, the Chancellor's Eminent Professor, and Senior Advisor for Strategy and Innovation. Before that, I was head of the Department of Computer Science, Edward William and Jane Marr Gutgsell Professor, and Director of the National Center for Supercomputing Applications (NCSA) at the University of Illinois.
I have served as a member of the President's Council of Advisors on Science and Technology (PCAST) and chair of the Computing Research Association (CRA).
See http://www.microsoft.com/presspass/exec/Reed/ for details on me at Microsoft.
~~~~~~~~~~~~~~~~~~~~~~~~
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>
]]>
By: Alexander Gray, AssociateProfessor
Computational Science and Engineering, College ofComputing, Georgia Tech
Date: Friday, September 23, 2011
Time: 2:00pm - 3:30pm
Location:Klaus 2447
Formore information please contact Dr. Alex Gray at agray@cc.gatech.edu
Title:
Techniques for Massive-Data Machine Learning
Abstract:
Startingwith motivations from data analysis problems in astronomy as examples, we'llconsider the task of making state-of-the-art machine learning methodsscale to massive datasets (including n-point correlation functions, kerneldensity estimation, minimum spanning trees, bipartite matching, nonparametricBayes classifiers, support vector machines, Nadaraya-Watson regression, kernelconditional density estimation, Gaussian process regression, nearest-neighbors,principal component analysis, hierarchical clustering, and manifold learning),despite their often quadratic or cubic scaling with the number of data, viaseven different types of computational techniques: indexing, functionaltransforms, sampling, problem reductions, locality, parallelism, and activelearning.
Bio:
Alexander Gray received bachelor's degrees in Applied Mathematics and ComputerScience from the University of California, Berkeley and a PhD in ComputerScience from Carnegie Mellon University, and is currently an AssociateProfessor in the College of Computing at Georgia Tech. His group ofapproximately 20 researchers, the FASTlab, aims to comprehensively scale up allof the major practical methods of machine learning to massive datasets as wellas develop new statistical methodology and theory, guided by challenge problemsin cosmology, medicine, and other application areas. He began working withmassive scientific datasets in 1993 (long before the current fashionable talkof “big data”) at NASA's Jet Propulsion Laboratory in its Machine LearningSystems Group. High-profile applications of his large-scale ML algorithmshave been described in staff written articles in Science and Nature, includingcontributions to work selected by Science as the Top Scientific Breakthrough of2003. He has won or been nominated for a number of best paper awards instatistics and data mining and is a recipient of the National ScienceFoundation CAREER Award in 2009. He gives invited tutorial lectures onmassive-scale data analysis at the top data analysis research conferences,government agencies, and corporations, and is a member of the prestigiousNational Academy of Sciences Committee on the Analysis of Massive Data.
]]>
By: Amihood Amir
Bar Ilan University and Johns Hopkins University
Date: Friday, September 9, 2011
Time: 2:00pm - 3:30pm
Location: Klaus 2447
For more information please contact Dr. Alberto Apostolico at axa@cc.gatech.edu
Title:
Cycle Detection and Correction
Abstract:
Assume that a natural cyclic phenomenon has been measured, but the data is corrupted by errors. The type of corruption is application-dependent and may be caused by measurement errors, or natural features of the phenomenon. We assume that an appropriate metric exists, which measures the amount of corruption experienced. We study the problem of recovering the corrupted cycle under various error models, formally called the period recovery problem. Specifically, we identify a metric property which we call pseudo-locality and study the period recovery problem under pseudo-local metrics. Examples of pseudo-local metrics are the Hamming distance, the swap distance, and the interchange (or Cayley) distance. We show that for pseudo-local metrics, periodicity is a powerful property allowing detecting the original cycle and correcting the data, under suitable conditions. Some surprising features of our algorithm are that we can efficiently identify the corrupted period, up to number of possibilities logarithmic in the length of the data string, even for metrics whose calculation is NP-hard.
Joint work with Estrella Eisenberg, Avivit Levy, Ely Porat, and Natalie Shapira
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>
Title:
Graphical Models for the Internet
Abstract:
In this talk I will present algorithms for performing large scale inference using Latent Dirichlet Allocation and a novel Cluster-Topic model to estimate user preferences and to group stories into coherent, topically consistent storylines. I will discuss both the statistical modeling challenges involved and the very large scale implementation of such models which allows us to perform estimation on over 50 million users on a Hadoop cluster.
Bio:
I studied physics in Munich at the University of Technology, Munich, at the Universita degli Studi di Pavia and at AT&T Research in Holmdel. During this time I was at the Maximilianeum München and the Collegio Ghislieri in Pavia. In 1996 I received the Master degree at the University of Technology, Munich and in 1998 the Doctoral Degree in computer science at the University of Technology Berlin. Until 1999 I was a researcher at the IDA Group of the GMD Institute for Software Engineering and Computer Architecture in Berlin (now part of the Fraunhofer Geselschaft). After that, I worked as a Researcher and Group Leader at the Research School for Information Sciences and Engineering of the Australian National University. >From 2004 onwards I worked as a Senior Principal Researcher and Program Leader at the Statistical Machine Learning Program at NICTA.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Title:
Non-parametric Bayes and Machine Learning
Bio:
Michael I. Jordan is the Pehong Chen Distinguished Professor in the Department of Electrical Engineering and Computer Science and the Department of Statistics at the University of California, Berkeley.
He was a professor at MIT from 1988 to 1998. His research in recent years has focused on Bayesian nonparametric analysis, graphical models and spectral methods, and applications to problems in signal processing, computational biology, and natural language processing. Prof. Jordan was named to the National Academy of Sciences in 2010 and the National Academy of Engineering in 2010.
He is a Fellow of the American Association for the Advancement of Science and a Fellow of the IMS, the ACM, and the IEEE.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Carnegie Mellon University
Title:
Mining Billion-node Graphs
Abstract:
What do graphs look like? How do they evolve over time? How to handle a graph with a billion nodes? We present a comprehensive list of static and temporal laws, and some recent observations on real graphs (like, e.g., "eigenSpokes"). For generators, we describe some recent ones, which naturally match all of the known properties of real graphs. Finally, for tools, we present "oddBall" for iscovering anomalies and patterns, as well as an overview of the PEGASUS system which is designed for handling Billion-node graphs, running on top of the "hadoop" system.
Bio:
Christos Faloutsos is a Professor at Carnegie Mellon University. He has received the Presidential Young Investigator Award by the National Science Foundation (1989), the Research Contributions Award in ICDM 2006, the SIGKDD Innovations Award (2010), seventeen "best paper" awards, (including two "test of time") and four teaching awards. He has served as a member of the executive committee of SIGKDD; he is an ACM Fellow; he has published over 200 refereed articles, 11 book chapters and one monograph. He holds five patents and he has given over 30 tutorials and over 10 invited istinguished lectures. His research interests include data mining for graphs and streams, fractals, database performance, and indexing for multimedia and bio-informatics data.
]]>University of California Santa Barbara
Title:
Spatial Stochastic Simulation of Polarization in Yeast Mating
Abstract:
In microscopic systems formed by living cells, the small numbers of some reactant molecules can result in dynamical behavior that is discrete and stochastic rather than continuous and deterministic. Spatio-temporal gradients and patterns play an important role in many of these systems. In this lecture we report on recent progress in the development of computational methods and software for spatial stochastic simulation. Then we describe a spatial stochastic model of polarisome formation in mating yeast. The new model is built on simple mechanistic components, but is able to achieve a highly polarized phenotype with a relatively shallow input gradient, and to track movement in the gradient. The spatial stochastic simulations are able to reproduce experimental observations to an extent that is not possible with deterministic simulation.
Bio:
Dr. Linda Petzold is currently Professor in the Department of Computer Science (Chair 2003-2007) and the Department of Mechanical Engineering, and Director of the Computational Science and Engineering Program at the University of California Santa Barbara. She received her Ph.D. in Computer Science in 1978 from the University of Illinois. From 1978-1985 she was a member of the Applied Mathematics Group at Sandia National Laboratories in Livermore, California, from 1985-1991 she was Group Leader of the Numerical Mathematics Group at Lawrence Livermore National Laboratory, and from 1991-1997 she was Professor in the Department of Computer Science at the University of Minnesota. Dr. Petzold is a member of the US National Academy of Engineering. She was recently named the Faculty Research Lecturer for 2011, the highest honor that UCSB bestows to its faculty. She is a Fellow of the ASME,SIAM and AAAS. She was awarded the Wilkinson Prize for Numerical Software in 1991, the Dahlquist Prize in 1999, and the AWM/SIAM Sonia Kovalevski Prize in 2003. She served as SIAM (Society for Industrial and Applied Mathematics) Vice President at Large from 2000-2001, as SIAM Vice President for Publications from 1993-1998, and as Editor in Chief of the SIAM Journal on Scientific Computing from 1989-1993.
]]>Post-doctoral Fellow at School of Computer Science, Carnegie Mellon University
Title:
Modeling Rich Structured Data via Kernel Distribution Embeddings
Abstract:
Real world applications often produce a large volume of highly uncertain and complex data. Many of them have rich microscopic structures where each variable can take values on manifolds (e.g., camera rotations), combinatorial objects (e.g., texts, graphs of drug compounds) or high dimensional continuous domains (e.g., images and videos). Furthermore, these problems may possess additional macroscopic structures where the large collections of observed and hidden variables are connected by networks of conditional independence relations (e.g., in predicting depth from still images, and forecasting in time-series).
Most previous learning algorithms for problems with such rich structures rely heavily on linear relations and parametric models where data are typically assumed to be multivariate Gaussian or discrete with a relatively small number of values. Conclusions inferred under these restricted assumptions can be misleading, if the underlying data generating processes contain nonlinear, non-discrete, or non-Gaussian components.
How can we find a suitable representation for nonlinear and non-Gaussian relationships in a data-driven fashion? How can we exploit conditional independence structures between variables in rich structured setting? How can we design efficient algorithms to solve challenging nonparametric problems involving large amount of data?
In this talk, I will introduce a nonparametric representation for distributions called kernel embeddings that are capable of addressing problems with both microscopic and macroscopic structures. The key idea of the method is to map distributions to their expected features (potentially infinite dimensional), and given evidence, update these new representations solely in the feature space. Compared to existing nonparametric representations which are largely restricted to vectorial data and usually lead to intractable algorithms, very often kernel distribution embeddings lead to simpler, faster and more accurate algorithms in a diverse range of problems such as organizing photo albums, understanding social networks, retrieving documents across languages, predicting depth from still images and forecasting sensor time-series.
Bio:
Le Song is a post doctoral fellow at School of Computer Science, Carnegie Mellon University, working with a number of professors, including Eric Xing, Carlos Guestrin, Geoff Gordon and Jeff Schneider. Prior to that, Le Song obtained his PhD. degree in computer science from University of Sydney and National ICT Australia in 2008 under the supervision of Alex Smola. Le Song conducted research in statistical machine learning and data mining, with primary focus on kernel methods, probabilistic graphical models, and network analysis. He is also interested in large-scale machine learning problems, and machine learning applications in computational biology, texts, images, and network analysis.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Title:
Parallel Computing -- A Theoretical Perspective
Abstract:
The speeds of microprocessors are not increasing anymore. Yet the transistor sizes keep shrinking exponentially according to Moore's Law. This resulted in the paradigm shift from sequential to parallel computing: multi-cores processors -- processors with multiple CPUs -- have become a norm, rather than exceptions. Yet, we do not have good theoretical foundation for modern parallel systems and a clear understanding of what will make algorithms efficient on these systems.
In this presentation I will talk about the current landscape of modeling parallelism in modern multi-core architectures from the algorithmic perspective. I will also touch on emerging computational frameworks, alternative to multi-cores, such as GPUs and Google's MapReduce and how they are changing the landscape of parallel and distributed computing.
Bio:
Nodari Sitchinava received Bachelor and Master degrees in Electrical Engineering and Computer Science from MIT and a PhD in Computer Science from UC Irvine. After graduating he accepted a postdoctoral appointment at the MADALGO Center at the CS department of Aarhus University where he is to this day. His research concentrates on developing accurate models of computation for modern parallel architectures and designing algorithms for them. In particular, his PhD dissertation concentrated on combining cache-efficiency with parallelism for multi-core architectures and on the development of a number of fundamental combinatorial, graph and geometric algorithms in the new model.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>
Title:
Statistical Learning Algorithms for Discovering Hidden Structures in Data
Abstract:
Statistical modeling and inference of high-dimensional data is a common, yet challenging task in many areas. To address this important issue, exploiting hidden structures in data has become an increasingly appealing strategy. In this talk, I will describe two sets of work in that direction. I will start by describing how to identify low-dimensional structures which preserve probabilistic relations between random variables. To this end, we develop and apply techniques from nonparametric statistics to assert statistical (conditional) independences. I will then describe how to exploit sparse structures which lead to economical and interpretable probabilistic models. For this purpose, we investigate and propose sparsity-inducing regularization which result in sparse and low-rank solutions. Through out the talk, I will illustrate the utility of both types of structures with several application examples in visualization, pattern classification, exploratory data analysis, etc.
This talk is based on joint work with my students and other collaborators, under the support by NSF, DARPA and Google.
Bio:
Fei Sha is an assistant professor at the Computer Science Department, U. of Southern California. He obtained his doctoral degree in 2007 from U. of Pennsylvania. Before joining USC in 2008, he spent some time at UC Berkeley as a postdoc and Yahoo Research as a research scientist. His primary research interest has been theory and application of statistical machine learning.
]]>Assistant Professor at Purdue University in Computer Science and Statistics
Title:
Scalable Bayesian learning for complex and massive data
Abstract:
Computational analysis of complex data has become a driving force for scientific discovery and engineering applications. It is, however, often a challenging task due to the high dimensionality and the massive size of datasets. To address these challenges, we build sparse, relational, dynamic and nonparametric Bayesian models driven by various applications and develop efficient, scalable inference methods. In this talk, (i) I will describe a novel sparse Bayesian model that integrates generative and conditional models to select correlated variables, such as whole genome SNPs. (This model addresses the p>>n problem where p is the number of variables and n is the number of data points); (ii) I will present a Bayesian online learning algorithm that, unlike previous approaches, learns a dynamic compact representation of massive data and make predictions accordingly (the n>>p problem); And (iii) I will describe a parallel Bayesian inference method on graphics processing units to extract latent topic and clusters from data with both a large number of variables and samples (both n and p are large). In addition, I will present applications of these works, for example, in identifying genetic variations and biomarkers for the early diagnosis of Alzheimer’s disease, and modeling rare cell populations in flow cytometry data for the discovery of cancer stem cells.
Bio:
I obtained PhD from MIT in 2005 and worked as a postdoctoral researcher at MIT from 2005 to 2007. In 2007, I joined Purdue university as an Assistant Professor of Computer Science and Statistics. I received the A. Richard Newton Breakthrough Research Award from Microsoft Research in 2008, the Interdisciplinary Award from Purdue University in 2010, and the NSF CAREER award in 2011.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Scalable Bayesian learning for complex and massive data
]]>CANON Professor of Computer Science and Electrical Engineering at Stanford University
Title:
The Semiology of Graphics and Interactive Data Analysis: The Impact of Jacques Bertin on Information Visualization and Visual Analytics
Abstract:
In 1967 after years of experience as a cartographer, Jacques Bertin wrote a monumental book, the Semiology of Graphics, that describes a theory of graphical representations. In another book, Graphics and Graphic Information Processing, he described interative techniques for data analysis. These books have been very influential in the information visualization and visual analytics communities, despite being poorly understood. In memory of Bertin (1918-2010), I will review some of his major ideas, and describe how they have impacted the design of analysis and visualization systems such as Jock MacKinlay's Automatic Presentation System (APT) and Chris Stolte's Polaris/Tableau system.
Bio:
Pat Hanrahan is the CANON Professor of Computer Science and Electrical Engineering at Stanford University where he teaches computer graphics.
His current research involves visualization, image synthesis, virtual worlds, and graphics systems and architectures. Before joining Stanford he was a faculty member at Princeton.
Pat has also worked at Pixar where he developed developed volume rendering software and was the chief architect of the RenderMan(TM) Interface - a protocol that allows modeling programs to describe scenes to high quality rendering programs. In addition to PIXAR, he has founded two companies, Tableau and PeakStream, and served on the technical advisory boards of NVIDIA, Exluna, Neoptica, VSee and Procedural.
Professor Hanrahan has received three university teaching awards. He has received two Academy Awards for Science and Technology, the Spirit of America Creativity Award, the SIGGRAPH Computer Graphics Achievement Award, the SIGGRAPH Stephen A. Coons Award, and the IEEE Visualization Career Award. He was recently elected to the National Academy of Engineering and the American Academy of Arts and Sciences.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>On the best low multilinear rank approximation of higher-order tensors
Abstract:
Higher-order tensors are generalizations of vectors and matrices to third- or even higher-order arrays of numbers. In this talk, we consider a generalization of column and row matrix rank to tensors, called multilinear rank, and discuss the best low multilinear rank approximation problem. Given a higher-order tensor, we are looking for another tensor, as close as possible to the original one and with multilinear rank bounded by prespecified numbers. This approximation is used for dimensionality reduction and signal subspace estimation in higher-order statistics, biomedical signal processing, telecommunications and many other fields. We first consider two particular applications and then briefly present several algorithms for the computation of the approximation. Finally, we point out that the existence of local minima of the associated cost function could cause problems in real applications.
]]>Title
AM++: A Generalized Active Message Framework
Abstract
Active messages have been shown to be a good model to express irregular applications for distributed-memory systems. However, current active messaging frameworks are either too inflexible or too inefficient for use by graph algorithms. We have developed a new framework, AM++, for "generalized active messages" that provides both efficiency and flexibility. It also provides features such as flexible, user-configurable message coalescing and duplicate message elimination without users needing to entangle those behaviors into their applications. Implementation and benchmarking of graph algorithms using AM++ shows the usability benefits of the new features. I will also briefly discuss my work on the Graph 500 benchmark: the graph generators, MPI reference implementations, and the Argonne entry to the first Graph 500 list.
Bio
Jeremiah Willcock is a postdoctoral researcher in the Open Systems Lab at Indiana University. He received his Ph.D. from Indiana University in the area of generic compiler optimizations, then was a postdoctoral researcher in the ROSE project at Lawrence Livermore National Laboratory, in the area of program analysis. His current research interests include abstractions for high-performance and parallel computing, especially in the area of graph algorithms, and programming language features for generic programming.
Host
David Bader
Followed by questions with PIZZA and DRINKS
]]>Title:
"Algorithms for inference and experimental design in complex physical systems"
Abstract:
Simulation of complex physical systems increasingly rests on the interplay of experimental observations with computational models. Key inputs, parameters, or structural aspects of models may be incomplete or unknown, and must be developed from indirect and limited observations. At the same time, quantified uncertainties are needed to qualify computational predictions in the support of design and decision-making. In this context, Bayesian statistics provides a foundation for inference and for the optimal selection of experiments and observations. Computationally intensive models, however, can render a Bayesian approach prohibitive.
We will show that stochastic spectral methods, which have seen extensive development in the context of "forward" uncertainty propagation, are a useful tool for inference as well. We introduce a stochastic spectral formulation that accelerates Bayesian inference via rapid exploration of a surrogate posterior distribution. Theoretical convergence results are verified with several numerical examples---in particular, parameter estimation in transport processes and in chemical kinetic systems. We also extend this approach to high-dimensional and ill-posed inverse problems, estimating distributed quantities in a hierarchical Bayesian setting.
We will also discuss computational strategies for optimal experimental design---choosing experimental conditions to maximize information gain in parameters or outputs of interest. We propose a general Bayesian framework for experimental design with nonlinear simulation-based models, accounting for uncertainty in model parameters, experimental conditions, and observables. We then discuss efficient evaluation of the associated objective functions, coupled with stochastic optimization methods to maximize expected utility.
~~~~~~~~~~~~~~~~
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
The CSE Convocation is an opportunity to gather to celebrate and explore the scope of this new field of computing.
CSE is a collaborative discipline that synthesizes principles from mathematics, science, engineering and computing to create and apply computational models to solve today’s grand challenges. The School of CSE’s mission is to advance computational methods and techniques that create novel solutions to real world problems and enable discovery and innovation in science and engineering.
The Georgia Tech College of Computing has always been focused on transformation, and the School of CSE is at the forefront. Join us at the CSE Convocation as we examine the possibilities of this exciting field of study.
12:00 pm Lunch
12:40 pm Convocation Program (Klaus 1116)
1:00 pm Panel Session (Klaus 1116). Transformational Science: Past, Present, Future
2:00 pm Keynote Presentation: David Keyes (Klaus 1447)
King Abdullah University of Science and Technology
Dean, Mathematical & Computer Sciences & Engineering
Named Professor, Applied Mathematics & Computational Science
Dr. David Keyes is a pioneer in the development of large-scale simulation. He currently leads a mathematical cyberinfrastructure center for the U.S. Department of Energy under the Scientific Discovery through Advanced Computing (SciDAC) initiative, and was recognized with the Sidney Fernbach Award of the IEEE Computer Society (2007) and an ACM Gordon Bell Prize (1999). Keyes was named the Fu Foundation Professor of Applied Mathematics at Columbia University, the vice president of the Society for Industrial and Applied Mathematics (SIAM), and part-time director of the Institute for Scientific Computing Research at Lawrence Livermore National Laboratory (1999-2008).
Keyes began his career at Yale University, then joined Old Dominion University and the Institute for Computer Applications in Science & Engineering at NASA Langley Research Center (1993). He graduated summa cum laude with a BS in engineering in aerospace and mechanical sciences from Princeton University (1978), and earned a doctorate in applied mathematics from Harvard University (1984). He did postdoctoral work in the Computer Science Department of Yale University (1984-1985).
]]>Title:
Sequences of problems, matrices, and solutions
Abstract:
In a wide range of applications, we deal with long sequences of slowly changing matrices or large collections of related matrices and corresponding linear algebra problems. Such applications range from the optimal design of structures to acoustics and other parameterized systems, to inverse and parameter estimation problems in tomography and systems biology, to parameterization problems in computer graphics, and to the electronic structure of condensed matter. In many cases, we can reduce the total runtime significantly by taking into account how the problem changes and recycling judiciously selected results from previous computations. In this presentation, I will focus on solving linear systems, which is often the basis of other algorithms. I will introduce the basics of linear solvers and discuss relevant theory for the fast solution of sequences or collections of linear systems. I will demonstrate the results on several applications and discuss future research directions.
Bio:
Eric de Sturler received his PhD in 1994 at Delft University of Technology under the supervision of Henk van der Vorst. From 1993, he spent 5 years at ETH Zurich as (senior) research scientist at the Interdisciplinary Project Center for Supercomputing and the Swiss Center for Scientific Computing. He was a Leslie Fox prizewinner in 1997, and he spent the summer of 1997 visiting Stanford University at the invitation of Prof. Gene Golub. In 1998, he joined the faculty of the Department of Computer Science at the University of Illinois at Urbana-Champaign. He has been on faculty of the Mathematics Department at Virginia Tech since 2006. He was the Program Director of the SIAM Activity Group on Supercomputing from 2003 to 2006; he co-chaired the third SIAM Conference on Computational Science and Engineering in 2005; he is a general co-chair of the 13th IEEE International Conference on Computational Science and Engineering in 2010; and he is on the organizing committee of the SIAM Conference on Applied Linear Algebra in 2012. He has served as an editor for multiple journals, among others, for Applied Numerical Mathematics since 2005 and for SIAM Journal on Numerical Analysis from 2003 to 2009.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Associate Professor of Mathematics
Texas Tech University
Title:
Metanumerical computing for partial differential equations: the Sundance project
Abstract:
Metanumerical computing deals with computer programs that use abstract mathematical structure to manipulate, generate, and/or optimize compute-intensive numerical codes. This idea has gained popularity over the last decade in several areas of scientific computing, include numerical linear algebra, signal processing, and partial differential equations. The Sundance project is such an example, using high-level software-based differentiation of variational forms to automatically produce high-performance finite element implementations, all within a C++ library. In addition to automating the discretization of PDE by finite elements, recent work is demonstrating how to produce block-structured matrices and streamline the implementation of advanced numerical methods. I will conclude with some examples of this for some incompressible flow problems.
Bio:
Rob Kirby earned his Ph.D. in Computational and Applied Mathematics at the University of Texas at Austin in 2000. He worked as a Dickson Instructor and then Assistant Professor of Computer Science at the University of Chicago. In 2006, he moved to the Department of Mathematics & Statistics Department where he is currently an associate professor. His work has included analysis of numerical methods, automation and optimization of mathematical software, and low-complexity finite element algorithms on simplices.
http://www.math.ttu.edu/~kirby/
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>New Orleans is where the future began for renowned discoverers and innovators of the past, and in November 2010, The Big Easy plays host to discoverers of the future at SC10, the 23rd meeting in the conference's annual history.
It's a great match. No city in the world is better known for its celebrations and its ability to reinvent itself. And no conference rivals SC10's planned international lineup of celebrated speakers, panel participants, presentations, workshops and exhibits featuring the latest breakthroughs in high- performance computing, networking, storage and analysis.
Spotlighting the most original and fascinating scientific and technical applications from around the world, SC10 will bring together an unprecedented array of scientists, engineers, researchers, educators, programmers, system administrators, developers and program managers and an exceptional program of technical papers, tutorials and timely research posters. SC10's Exhibition Hall will be second to none, featuring exhibits from international participants representing industry, academia and government research organizations.
Mark your calendar and make your way to New Orleans - be there for SC10, November 13-19. No city dishes out a bigger helping of southern hospitality, and no conference offers a better opportunity not just to glimpse the future of computing, but to participate in it-and collaborate with its discoverers.
]]>Title:
Gravity's Strongest Grip: A Computational Challenge
Abstract:
Gravitational physics is entering a new era driven by observation that will begin once gravitational-wave interferometers make their first detections. In the universe, gravitational waves are produced during violent events such as the merger of two black holes. The detection of these waves, sometimes called ripples in the fabric of spacetime, is a formidable undertaking, requiring innovative engineering, powerful data analysis tools and careful theoretical modeling. High performance computing plays a vital role in our ability to predict and interpret gravitational waves with theoretical modeling of the sources. I will provide an overview of the high performance and data analysis challenges we face in making the first and subsequent detection of gravitational waves.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
Compressive Sensing with Applications to Fast Forward Modeling and Acoustic Source Localization
Abstract:
In the first part of this talk, we will give an overview of the recently developed theory of compressive sampling (CS), and discuss how this theory has influenced next-generation sensor design. The CS paradigm can be summarized neatly: the number of measurements (e.g. samples) needed to acquire a signal or image depends more on its inherent information content than on the desired resolution (e.g. number of pixels). The theory of CS is far-reaching and draws on subjects as varied as sampling theory, convex optimization, source and channel coding, statistical estimation, uncertainty principles, and harmonic analysis. The applications of CS range from the familiar (imaging in medicine and radar, high-speed analog-to-digital conversion, and super-resolution) to truly novel image acquisition and encoding techniques.
In the second part of the talk, we will show how some of the key ideas of CS can be used to dramatically reduce the computation required for two types of forward modeling problems. In the first, we show how all of the channels in a multiple-input multiple-output system can be acquired jointly by simultaneously exciting all of the inputs with different random waveforms, and give an immediate application to seismic forward modeling. In the second, we consider the problem of acoustic source localization in a complicated channel. We show that the amount of computation to perform "matched field processing" (matched filtering) can be reduced by precomputing the response of the channel to a small number of dense configurations of random sources.
]]>Title:
Novel Applications of Graph Embedding Techniques
Abstract:
Force-directed graph embedding algorithms, like the Fruchterman-Reingold method, are typically used to generate aesthetically pleasing graph layouts. At a fundamental level, these algorithms are based on manipulating the structural properties of the graph to match them to certain spatial requirements. This relation between structural and spatial properties is also present in other areas beyond graph visualization. In this talk, I will discuss how graph embedding can be used in diverse areas such as (i) improving the accuracy of unsupervised clustering, (ii) creating good quality elements in unstructured meshes and (iii) identifying perturbations in large-scale networks.
Bio:
Dr. Sanjukta Bhowmick is an Assistant Professor in the College of Information Science and Technology at the University of Nebraska at Omaha. She received her Bachelor of Technology degree from the Haldia Institue of Technology, West Bengal, India and her Ph.D. in Computer Science (with a minor in High Performance Computing) from the Pennsylvania State University and after her Ph.D. held a joint postdoctoral research position between Columbia University and Argonne National Laboratory. Dr. Bhowmick’s core research area involves the development of high performance computing algorithms for large scale combinatorial and numerical problems arising in varied applications such as social networks, computational fluid dynamics, datamining and bioinformatics. Her recent research includes large scale data analysis using combinatorial algorithms such as force-directed graph embedding methods and parallel algorithms for community detection in large dynamic networks.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
The Joy of PCA
Santosh Vempala (Georgia Tech)
Principal Component Analysis is the most widely usedtechnique for high-dimensional or large data. For typical applications (nearestneighbor, clustering, learning), it is not hard to build examples on which PCA*fails*. Yet, it is popular and successful across a variety of data-rich areas.In this talk, we focus on two algorithmic problems where the performance of PCAis provably near-optimal, and no other method is known to have similarguarantees. The problems we consider are (a) the classical statistical problemof unraveling a sample from a mixture of k unknown Gaussians and (b) theclassic learning theory problem of learning an intersection of k halfspaces.During the talk, we will encounter recent extensions of PCA that arenoise-resistant, affine-invariant and nonviolent.
]]>Digital image processing is playing an important role in all fields of Science and Technology today. Pictures/picture signals are digitized with 8-bit to 16-bit digitizers to accommodate vide dynamic range of picture signals. Pictures are computer processed for image enhancement, geometric and radiometric errors, etc., Registration of temporal images are carried out for checking the changes in the images with time, of a patient in case of health care, remote sensing images in case of crop monitoring, etc., Image processing techniques so far developed for processing satellite images ,images from space probes are being adopted for healthcare applications.
Imaging techniques are constantly developed in medicine using different modality devices such as MRI, CT, Ultra-sound and PET. 2D images are used to analyze the diseases, while some of these images are combined to generate 3D volumes of human body for calculating volume of affected areas.
Various hyper spectral imaging techniques, which map wavelength-resolved reflectivity across a two dimensional scene offer considerable promise as a new clinical tool for ophthalmology. Due to the delicate nature of the eye, invasive biopsy or mechanical access to the retina are not possible and relies strongly upon optical imaging methods for diagnosis of diseases connected to Carnia, retina and optic disc.
Lot of technologies such as image processing, hyper spectral imaging, networking, etc., developed for various applications, if adopted with suitable modifications for health care, will benefit millions of people for getting healthcare at low cost.
This talk covers Satellite Image processing, Remote sensing sensors and use of image processing for Healthcare.
Dr.K.M.M.Rao received his B.E (Honors) and M.E. in Electronics from Birla Institute of Technology and Science, Pilani in 1970 & 1972, respectively and his Ph.D in Image Processing from Sri Krishnadevaraya University, Andhra Pradesh (INDIA).
Dr.K.M.M.Rao joined Indian Institute of Science(I.I.Sc), Bangalore in 1973. He was responsible for the design and development of Black and White and Colour Drum Scanner Imager, first time in India, for digitizing and reproducing various types of photographic prints. At Institute of Science, he carried out research in Image Processing, Aerial Remote Sensing and Photo writing. Since 1976, he worked with National Remote Sensing Center (NRSC),ISRO in various positions. He was designated as Operations Director IRS-1C in 1996 and IRS-1D in 1997. He was Deputy Director (Data Processing Area) at NRSC, till December 2006. Currently, he is Adjunct Professor BITS-Pilani associated with Hyderabad campus.
He has designed and developed number of technologies for Image Processing and Remote Sensing applications. He obtained five patents for his designs.
He has designed and developed software to analyze planimetric parameters of optic disk to study glaucoma disease objectively. He developed software to montage fundus images to study retinal diseases.He has developed multimodal image registration and fusion techniques for medical images. He has contributed for improving the quality of radiographic images and to develop computer hardware for digital catheterization laboratory. He works with leading Hospitals in India to validate software and hardware designs.
He has won the National Academy of Science India Award along with a gold medal, citation and cash award, in the field of Instrumentation in 1995,Drum Scanner-Digitizer bagged the Best Invention Award in 1994,Presidential Award for Meritorious Invention of Color Photowrite System on 15th August, 1993. Best Invention Award for Color Photo write System in 1992,Engineer of the Year Award from Institute of Engineers, AP State Centre for 1989,Satellite Image Processing System (SIPS) bagged Best Invention Award in 1989,etc.,
He is Adviser to number of companies in India. He is Fellow of Institute of Electronics & Telecommunications (IETE) India, Fellow of ISNT, Senior Member IEEE (US), Life Member Indian Society of Remote Sensing (ISRS). He was a Member in Board of Studies in number of Universities.
He has more than 120 research publications/reports to his credit. He has supervised over 200 graduate and undergraduate students in their research projects. He is currently supervising few Ph.D and M.S students. Recently Two of his students got Ph.D from JNTU, Hyderabad.
He was Guest editor for International Journal of Applied Earth Observation and Geoinformation on Indian Remote Sensing Satellite, Resourcesat-1. He has visited number of countries, such as USA, Canada, Netherlands, Singapore, Dubai, etc.
His areas of interest include Satellite Data Processing, Image Processing, Medical Imaging, Data Mining and Photo Writing.
]]>Refreshments will be served following the address.
]]>"Playing repeated games: Theory, an algorithm, applications"
Abstract:
This talk will describe a simple, general algorithm for learning to play any matrix game against an unknown adversary. The algorithm, which is based directly on Littlestone and Warmuth's weighted majority algorithm, can be shown never to perform much worse than the best fixed strategy, even if selected in hindsight. Moreover, because of the algorithm's moderate resource requirements, it can be used even when working with extremely large game matrices. Taken together, these properties make the algorithm a good fit for a range of machine-learning applications, including on-line learning and boosting. Recently, the algorithm has also been applied to reinforcement learning, specifically, to the problem of learning to imitate the behavior of an "expert" while attempting simultaneously to improve on the expert's performance.
Bio:
Robert Schapire received his ScB in math and computer science from Brown University in 1986, and his SM (1988) and PhD (1991) from MIT under the supervision of Ronald Rivest. After a short post-doc at Harvard, he joined the technical staff at AT&T Labs (formerly AT&T Bell Laboratories) in 1991 where he remained for eleven years. At the end of 2002, he became a Professor of Computer Science at Princeton University. His awards include the 1991 ACM Doctoral Dissertation Award, the 2003 Gödel Prize and the 2004 Kanelakkis Theory and Practice Award (both of the last two with Yoav Freund). His main research interest is in theoretical and applied machine learning.
]]>"Nonparametric Graphical Models"
Abstract:
Graphical modeling has proven to be an extremely useful abstraction in statistical machine learning. The space of possible graphical models is enormous, yet only a very limited set of models has been extensively developed for continuous data. The most basic, classical example is the Gaussian graphical model, where the precision matrix encodes the independence graph. While Gaussian graphical models can be useful, a reliance on exact normality is limiting. We present recent work for estimating nonparametric graphical models. One approach is something we call "the nonparanormal," which uses copula methods to transform the variables by nonparametric functions, relaxing the strong distributional assumptions made by the Gaussian graphical model. Another approach is to restrict the family of allowed graphs to spanning forests, enabling the use of fully nonparametric density estimation. The resulting methods are easy to understand, simple to use, theoretically well supported, and effective for modeling of high dimensional data. Joint work with Anupam Gupta, Han Liu, Larry Wasserman, and Min Xu.
Bio:
John Lafferty is a professor in the Computer Science Department and the Machine Learning Department within the School of Computer Science at Carnegie Mellon University, where he also holds a joint appointment in the Department of Statistics. His research interests are in text analysis, machine learning, and statistical learning theory, with a recent focus on theory and methods for high dimensional data.
Courtesy of our generous sponsor, Yahoo!
Free Pizza!
]]>
Over the last decades, much of the research on discriminative learning has focused on problems like classification and regression, where the prediction is a single univariate variable. But what if we need to predict complex objects like trees, orderings, or alignments? Such problems arise, for example, when a natural language parser needs to predict the correct parse tree for a given sentence, when one needs to optimize a multivariate performance measure like the F1-score, or when predicting the alignment between two proteins.
This talk discusses a support vector approach and algorithm for predicting such complex objects. It generalizes conventional classification SVMs to a large range of structured outputs and multivariate loss functions. While the resulting training problems have exponential size, there is a simple algorithm that allows training in polynomial time. The algorithm is implemented in the SVM-Struct software and empirical results will be given for several examples.
Thorsten Joachims is an Associate Professor in the Department of Computer Science at Cornell University. In 2001, he finished his dissertation with the title "The Maximum-Margin Approach to Learning Text Classifiers: Methods, Theory, and Algorithms", advised by Prof. Katharina Morik at the University of Dortmund. From there he also received his Diplom in Computer Science in 1997 with a thesis on WebWatcher, a browsing assistant for the Web. From 1994 to 1996 he was a visiting scientist at Carnegie Mellon University with Prof. Tom Mitchell. His research interests center on a synthesis of theory and system building in the field of machine learning, with a focus on Support Vector Machines and machine learning with text. He authored the SVM-Light algorithm and software for support vector learning.
]]>Abstract:
Classical multidimensional scaling (MDS) is a method for visualizing high-dimensional point clouds by mapping to low-dimensional Euclidean space. This mapping is defined in terms of eigenfunctions of a matrix of inter-point proximities. I will show cases where a complex phenomena can be broken down into simple one dimensional components using versions of MDS, one example is joint work with S. Goel and P. Diaconis on roll call votes from the 2005
house of representatives votes. Another is joint work with J. Chakerian and uses a distance between trees to see differences inhierarchical trees built from sequence data at varying mutation rates.
Bio:
Susan P. Holmes is a Professor of Statistics and Associate Director of the Mathematical and Computational Sciences Interdisciplinary Program. Dr. Holmes' main themes of research are: Computational Biology, Computer intensive methods in multivariate statistics, Phylogenetic analysis of DNA sequences, Multivariate statistics applied to micro-array techniques.
Refreshments will be provided.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
"3I’s of Visual Analytics for FODAVA Teams: Interdisciplinary, International, and Immediacy"
Abstract:
Visual Analytics is an emerging interdisciplinary field of study that brings talents from many disciplines including statistics, mathematics, information, knowledge, and library sciences, knowledge representation and synthesis, scientific and information visualization, cognitive and perceptual sciences, communications, decision sciences and more. The demand for visual analytics is being stimulated by new requirements for homeland security but similar needs are present in science, commerce, home, and almost any domain that deals with complex, large information sources that require human judgement to “detect the expected and discover the unexpected”. Almost all of these needs requires interculture international solutions with near term needs while keeping on eye on the long term science. The definition of visual analytics is the science of analytical reasoning facilitated by the interface visual interface. Jim will present the new needs for science and technology, referenced from the recent book Illuminating the Path: the Research and Development Agenda for Visual Analytics, http://nvac.pnl.gov/. Jim will also discuss success stories, common characteristics of todays visual analytics solutions, directly addressing the 3I’s, and provide Jim’s top technical challenges for visual analytics, enlisting comments and recommendations.
Bio:
Dr. Jim Thomas is an American Association for the Advancement of Science (AAAS) Fellow and Laboratory Fellow at Pacific Northwest National Laboratory with over 35 years of experience. In 2009, he was awarded Christopher Columbus Foundation Award for top science and impact in contributing to homeland security, and DHS Award for Accomplishment: Technology Partnership and Deployment. He is founder and past Director of Department of Homeland Security National Visualization and Analytics Center. He is considered the father for visual analytics and specializes in the research, design, and implementation of innovative information and analytic visualization, multimedia, and human computer interaction technology with over 160 published papers in these areas. Some of the recent technologies developed have set a new stage for the visualization of masses of multimedia information sources with several publications, patents, with recent publications being widely referenced and re-printed. More recently he has led teams in text, numerical, image and video, temporal and geospatial analysis for massive information spaces.
He has received several international science awards including "Top 100 Scientific Innovators" (Science Digest) and twice the Research and Development's Industrial Research 100 Significant Scientific and Industry Accomplishments "Top 100 Innovators in Science and Industry". In addition, twice he was awarded the Federal Laboratories Consortium Technology Transfer Award for innovation in transferring research technology to industry and universities. He has served numerous other leadership roles including Chair of IEEE Visual Analytics Science and Technology Symposium Steering Committee 2006-present, IEEE Visualization Conference Co-Chair 2003-2004, Editor-In-Chief for IEEE Computer Graphics and Applications 1998-2002 and Chair of ACM SIGGRAPH 1987 –1992.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>"Markov Logic Networks: A Language for Statistical Relational Learning"
Abstract:
Modern machine learning applications are characterized by high degrees of complexity and uncertainty. Complexity is well handled by first-order logic, and uncertainty by probabilistic graphical models. Statistical relational learning seeks to combine the two. Markov logic networks (MLNs) do this by attaching weights to logical formulas and treating them as templates for features of Markov random fields. This talk will cover MLN representation, inference, learning and applications. MLN inference techniques are based on satisfiability testing, resolution, Markov chain Monte Carlo, and belief propagation. Learning techniques include pseudo-likelihood, voted perceptrons, second-order convex optimization, and inductive logic programming. MLNs have been applied in a wide variety of areas, including natural language processing, information extraction and integration, robot mapping, social networks, computational biology, and others. Open-source implementations of MLN algorithms are available in the Alchemy package (alchemy.cs.washington.edu). (Joint work with Jesse Davis, Stanley Kok, Daniel Lowd, Aniruddh Nath, Hoifung Poon, Matt Richardson, Parag Singla, Marc Sumner, and Jue Wang.)
Bio:
Pedro Domingos is Associate Professor of Computer Science and Engineering at the University of Washington. His research interests are in artificial intelligence, machine learning and data mining. He received a PhD in Information and Computer Science from the University of California at Irvine, and is the author or co-author of over 150 technical publications. He is a member of the editorial board of the Machine Learning journal, co-founder of the International Machine Learning Society, and past associate editor of JAIR. He was program co-chair of KDD-2003 and SRL-2009, and has served on numerous program committees. He has received several awards, including a Sloan Fellowship, an NSF CAREER Award, a Fulbright Scholarship, an IBM Faculty Award, and best paper awards at KDD-98, KDD-99, PKDD-05 and EMNLP-09.
]]>Assistant Research Professor
Department of Electrical and Computer Engineering
Carnegie Mellon University
For more information please contact Dr. Richard Vuduc
Title:
Load-Balanced Bonded Force Calculations on Anton
Abstract:
Spiral (www.spiral.net) is a program and hardware design generation system for linear transforms such as the discrete Fourier transform, discrete cosine transforms, filters, and others. We are currently extending Spiral beyond its original problem domain, using coding algorithms (Viterbi decoding and JPEG 2000 encoding) and image formation synthetic aperture radar, SAR) as examples. For a user-selected problem specification, Spiral autonomously generates different algorithms, represented in a declarative form as mathematical formulas, and their implementations to find the best match to the given target platform. Besides the search, Spiral performs deterministic optimizations on the formula level, effectively restructuringthe code in ways unpractical at the code or design level. Spiral generates specialized single-size implementations or adaptive general-size autotuning libraries, and utilizes special instructions and multiple processor cores.
The implementation generated by Spiral rival the performance of expertly hand-tuned libraries.In this talk, we give a short overview on Spiral. We explain then howSpiral generates efficient programs for parallel platforms including vector architectures, shared and distributed memory platforms, and GPUs; as well as hardware designs (Verilog) and automatically partitioned software/hardware implementations. We overview how Spiral targets the Cell BE and PowerXCell 8i, the BlueGene/P PPC450d processors, as well as Intel's upcoming Larrabee GPU and AVX vector instruction set. As all optimizations in Spiral, parallelization and partitioning are performed on a high abstraction level of algorithm representation, using rewriting systems.
Bio:
Franz Franchetti is an Assistant Research Professor with the Department of Electrical and Computer Engineering at Carnegie Mellon University. He received the Dipl.-Ing. (M.Sc.) degree in Technical Mathematics and the Dr. techn. (Ph.D.) degree in Computational Mathematics from the Vienna University of Technology in 2000 and 2003, respectively. He was a postdoctoral research associate with the Institute for Analysis and Scientific Computing during 2003. In 2004-2005 he was a postdoctoral research associate with the Department of Electrical and Computer Engineering at Carnegie Mellon University and a recipient of the Schrödinger fellowship awarded by the Austrian Science Fund. In 2006 he was member of the team winning the Gordon Bell Prize (Peak Performance Award). From 2005-2008 he was Systems Scientist (special faculty) with Carnegie Mellon’s ECE Department.
Dr. Franchetti's research focuses on automatic performance tuning and program generation for emerging parallel platforms, including multicore CPUs, clusters and high-performance systems (HPC), graphics processors (GPUs), field programmable gate arrays (FPGAs), and FPGA-acceleration for CPUs. His research goal is to enable automatic generation of highly optimized software libraries for important kernel functionality. He is member of the Spiral research team at CMU (www.spiral.net) and co-founder of SpiralGen (www.spiralgen.com), which is commercializing the Spiral technology. More information can be found at http://www.ece.cmu.edu/~franzf.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Machine Learning Department at Carnegie Mellon
For more information please contact Dr. Guy Lebanon
Title:
Fast Algorithms for Querying and Mining Large Graphs
Abstract:
Graphs appear in a wide range of settings and have posed a wealth of fascinating problems. In this talk, I will present our recent work on (1) querying (e.g., given a social network, how to measure the closeness between two persons? how to track it over time?); and (2) mining (e.g., how to identify abnormal behaviors of computer networks? In the case of virus attacks, which nodes are the best to immunize?) large graphs.
For the task of querying, our main finding is that many complex user-specific patterns on large graphs can be answered by means of proximity measurement. In other words, proximity allows us to query large graphs on the atomic levels. Then, I will talk about how to adapt querying tasks to the time evolving graphs. For fast computation of proximity, we developed a family of fast solutions to compute the proximity in several different scenarios. By carefully leveraging some important properties shared by many real graphs (e.g., the block-wise structure, the linear correlation, the skewness of real bipartite graphs, etc), we can often achieve orders of magnitude of speedup with little or no quality loss. For the task of mining, I will talk about immunization and anomaly detection. For immunization, we proposed a near-optimal, fast and scalable algorithm. For anomaly detection, we proposed a family of example-based low-rank matrix approximation methods. The proposed algorithms are provably equal to or better than best known methods in both space and time, with the same accuracy. On real data sets, it is up to 112x faster than the best competitors, for the same accuracy.
Bio:
Hanghang Tong got his Ph.D in the Machine Learning Department at Carnegie Mellon University in 2009. He has received best paper awards from SIAM-DM 2008 and ICDM 2006. He holds an M.S. degree and a B.S. degree from Tsinghua University, P.R. China. His research interests include data mining for multimedia.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Distinguished Lecture in Bioinformatics
Dr. Gill Bejerano
Developmental Biology and Computer Science Stanford University
Contact: Mark Borodovsky
Title:
Conservation and function in the human genome
Abstract:
I will present a modern view of the human genome, particularly as it relates to gene regulation, which appears to be encoded by a vastly more complex genomic layer than we ever suspected. I will describe some of our own work that helps us appreciate the complexity and mysteries that make our genome such a fascinating research topic.
]]>"Statistical Inference using Weak Chaos and Infinite Memory"
Abstract:
We describe a class of deterministic weakly chaotic dynamical systems with infinite memory. These ``herding systems'' combine learning and inference into one algorithm, where moments or data-items are converted directly into an arbitrarily long sequence of pseudo-samples. This sequence has infinite range correlations and as such is highly structured. We show that its information content, as measured by sub-extensive entropy, can grow as fast as K log(N), which is faster than the usual 1/2 K log(N) for exchangeable sequences generated by random posterior sampling from a Bayesian model. In continuous spaces we show that a kernel version of herding generates samples form a density that, when used to compute Monte Carlo sums, converges at a rate O(1/T) (as opposed to O(1/sqrt(T)) for random samples). More generally, we advocate the application of the rich theoretical framework of nonlinear dynamical systems, chaos theory and fractal geometry to statistical learning.
Bio:
Max Welling is a Professor of Computer Science at UC Irvine with a joint appointment in the statistics department. He is associate director of the Center for Machine Learning and Intelligent Systems, associate editor for TPAMI and JCGS journals. He received multiple grants from NSF, NIH and ONR-MURI among which an NSF career grant in 2005. He was recipient of the Dean’s midcareer award for research in 2008. He was conference chair for the Conference on AI and Statistics in 2009. Before joining UCI he held postdoctoral positions at Caltech (’98-’00), University College London (’00-’01) and the University of Toronto (’01-’03). He received his PhD in ’98 in theoretical physics.
His research focuses on large scale statistical learning. He has made contributions in approximate inference in graphical models, hierarchical models of complex cells, products of expert models, algorithms for learning image taxonomies, visual object recognition, information retrieval, text models, image denoising, and statistical shape analysis. He has over 70 publications in machine learning.
You are cordially invited to attend a reception in the lounge next to Klaus 1324 before the seminar to chat informally with faculty and students. PIZZA will be provided.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
Parallel computing is entering the era of petascalemachines. This era brings enormous computing power to us and newchallenges to harness this power efficiently. Machines with hundreds ofthousands of processors already exist, connected by complexinterconnect topologies. Network contention is becoming an increasinglyimportant factor affecting overall performance. The farther differentmessages travel on the network, greater is the chance of resourcesharing between messages and hence, of contention. Recent studies onIBM Blue Gene and Cray XT machines have shown that under contention,message latencies can be severely affected.
Mapping ofcommunicating tasks on nearby processors can minimize contention andlead to better application performance. In this talk, I will proposealgorithms and techniques for automatic mapping of parallelapplications to relieve the application developers of this burden. Iwill first demonstrate the effect of contention on message latenciesand use these studies to guide the design of mapping algorithms. I willintroduce the hop-bytes metric for the evaluation of mapping algorithmsand suggest that it is a better metric than the previously used maximumdilation metric. I will then discuss in some detail, the mappingframework which comprises of topology aware mapping algorithms forparallel applications with regular and irregular communication patterns.
Iwill also briefly discuss my interests within and future ideas forparallel computing research. More details on my research available at:http://charm.cs.illinois.edu/~bhatele/phd/
Abhinavreceived a B. Tech. degree in Computer Science and Engineering fromI.I.T. Kanpur (INDIA) in May 2005 and a M. S. degree in ComputerScience from the University of Illinois at Urbana-Champaign in 2007. Heis a 5th year Ph.D. student at the Parallel Programming Lab at theUniversity of Illinois, working with Prof. Laxmikant V. Kale. Hisresearch is centered around topology aware mapping and load balancingfor parallel applications. Abhinav has received the David J. KuckOutstanding MS Thesis Award in 2009, Third Prize in the ACM StudentResearch Competition at SC 2008, a Distinguished Paper Award atEuro-Par 2009 and the George Michael HPC Fellowship Award at SC 2009.
]]>The Mathematical Foundation of Analytic Visualizations
The Grammar of Graphics (GoG) is the title of a book that lays out the mathematical foundation of analytic visualizations: statistical, cartographic, and other quantitative graphics designed to represent observed or abstract data. Analytic visualizations are distinguished from other graphics by their mathematical formalism. Informal diagrams, by contrast, are designed to communicate ideological, artistic, religious, or other metaphorical information.
The GoG foundation is based on the conventional definition of the graph of a function: a collection of ordered pairs (x, f(x)). A graphic is a visual representation of the graph of a function. In analytic visualizations, this function operates on observed or abstract data.
GoG decomposes the global visualization function into seven orthogonal classes that comprise a totally ordered function chain. Each class has a collection of member functions that are composable with functions in adjacent classes of the function chain. The first class (Variables) maps data to an object called a varset (a set of variables). The next two classes (Algebra, Scales) are transformations on varsets. The next class (Statistics) takes a varset and creates a statistical graph (a statistical summary). The next class (Geometry) maps a statistical graph to a geometric graph. The next (Coordinates) embeds a graph in a coordinate space. And the last class (Aesthetics) maps a graph to a visible or perceivable display called a graphic.
A consequence of this class-orthogonality is a high degree of expressiveness: the product set of these seven function classes produces a huge variety of graphical forms or chart types. In fact, it is claimed that virtually the entire corpus of known statistical charts can be generated by this relatively parsimonious system, and perhaps a great number of meaningful but undiscovered chart types as well.
The second principal claim of GoG is that this function chain encapsulates the meaning of what we do when we construct formal statistical graphics, charts, and visualizations. It is not a taxonomy. It is a computational system based on the underlying mathematics of
representing functions of data. A consequence of this claim is to say that charts not definable within the GoG chain should be carefully examined for the possibility that they are ill-formed (meaningless).
This talk will include concrete examples to illustrate distinguishing characteristics of visualization languages based on GoG: simplicity, expressiveness, coherence, and meaningfulness. I will also survey software systems based on GoG that have been developed since the book was first published in 1999.
Leland Wilkinson is Executive VP of SYSTAT Software Inc., Adjunct Professor of Statistics at Northwestern University, and Adjunct Professor of Computer Science at the University of Illinois Chicago. He received an A.B. degree from Harvard in 1966, an S.T.B. degree from Harvard Divinity School in 1969, and a Ph.D. from Yale in 1975. Wilkinson wrote the SYSTAT statistical package and founded SYSTAT Inc. in 1984. After the company grew to 50 employees, he sold SYSTAT to SPSS in 1994 and worked there for ten years on research and development of visualization systems. SPSS eventually sold SYSTAT to Cranes Software International and Wilkinson rejoined SYSTAT in 2008.
Wilkinson is a Fellow of the American Statistical Association, an elected member of the International Statistical Institute, and a Fellow of the American Association for the Advancement of Science. He has won best speaker award at the National Computer Graphics Association and the Youden prize for best expository paper in the statistics journal Technometrics. He has served on the Committee on Applied and Theoretical Statistics of the National Research Council and has been Vice Chair of the Board of the National Institute of Statistical Sciences (NISS). In addition to authoring journal articles, the original SYSTAT computer program and manuals, and patents in visualization and distributed analytic computing, Wilkinson is the author (with Grant Blank and Chris Gruber) of Desktop Data Analysis with SYSTAT and The Grammar of Graphics.
]]>"What do other people think?" has always been an important consideration to most of us when making decisions. Long before the World Wide Web, we asked our friends who they were planning to vote for and consulted Consumer Reports to decide which dishwasher to buy. But the Internet has (among other things) made it possible to learn about the opinions and experiences of those in the vast pool of people that are neither our personal acquaintances nor well-known professional critics --- that is, people we have never heard of. Enter sentiment analysis, a flourishing research area devoted to the computational treatment of subjective and opinion-oriented language. Sample phenomena to contend with range from sarcasm in blog postings to the interpretation of political speeches. This talk will cover some of the motivations, challenges, and approaches in this broad and exciting field.
Lillian Lee is a professor of computer science at Cornell University. Her research interests include natural language processing, information retrieval, and machine learning. She is the recipient of the inaugural Best Paper Award at HLT-NAACL 2004 (joint with Regina Barzilay), a citation in "Top Picks: Technology Research Advances of 2004" by Technology Research News (also joint with Regina Barzilay), and an Alfred P. Sloan Research Fellowship, and her group'swork has been featured in the New York Times.
]]>
Beginning with a brief overview of a boundary integral problem that led from an intractable computational problem to model an efficient solar cell but through a novel mathematical formulation became computationally tractable, I will lay the foundation of the importance of mathematics in computation. This problem led to my career in Computational Science and High Performance Computing (HPC). Today, as we move from the Petascale era to Exascale era, HPC is a tool frequently used to understand complex problems in numerous areas such as aerospace, biology, climate modeling and energy. Scientists and engineers working on problems in these and other areas demand ever increasing compute power for their problems. In order to satisfy the demand for increase performance to achieve breakthrough science and engineering, we turn to parallelism through large systems with multi-core chips. For these systems to be useful massive parallelism at the chip level is not sufficient. I will describe some of the challenges that will need to be considered in designing Petascale and eventually Exascale systems. Through the combination of HPC hardware coupled with novel mathematical and algorithmic approaches, such as those described in the initial problem of this talk, some efforts toward breakthroughs in science and engineering are described. While progress is being made, there remain many challenges for the computational science and engineering community to apply ultra-scale, multi-core systems to “Big” science problems with impact on society. New radical algorithmic approaches are needed that emphasize minimization of data transmission instead of floating point performance. In conclusion, some discussion not only on the most obvious way to use ultra-scale, multi-core HPC systems will be given but also some thoughts on how one might use such systems coupled with new computational science techniques to tackle previously intractable problems.
Dr. Kirk E. Jordan, Emerging Solutions Executive in the Computational Science Center at IBM T.J. Watson Research Center, has more than 25 years experience in high performance and parallel computing. The Computational Science Center is addressing the challenges involved in achieving Petascale and Exascale performance on IBM’s very high end system platforms, running real workloads to obtain significant results in science, engineering, business and social policy, and partnering and collaborating with key IBM clients on the most challenging applications and workloads on these large systems. He oversees development of applications for IBM’s advanced computing architectures, investigates and develops concepts for new areas of growth involving high performance computing (HPC), and provides leadership in high-end computing and simulation in such areas as computational fluid dynamics, systems biology and high-end visualization. At IBM, he held several positions promoting HPC and high performance visualization, including leading technical efforts in the Deep Computing organization within IBM’s Systems and Technology Group, managing IBM’s University Relations SUR(SharedUniversity Research) Program and leading IBM’s Healthcare and Life Sciences Strategic Relationships and Institutes of Innovation Programs. A Ph.D. in Applied Math, he held computational science positions at Exxon R&E, Argonne National Lab, Thinking Machines and Kendall Square Research before joining IBM in 1994. A Research Affiliate in MIT’s Department of Aeronautic and Astronautics, he holds leadership positions in the Society for Industrial and Applied Mathematics (SIAM), including Vice Chair of Computational Science and Engineering SIAG and the Committee on Science Policy. He is on several boards including Math Biosciences Institute’s Board of Trustees at The Ohio State University, Board of the National Professional Sciences Master’s Association, and the International Advisory Board for the Systems Biomedicine Institute at Shanghai Jaio Tong University. He is associate editor of several international journals and Guest Editor for two recent issues of IBM’s Journal for Research and Development.
]]>"The tau-path test for monotone association in an unspecified subpopulation: Application to chemogenomic data mining"
Abstract:
In data mining and other settings, there is sometimes a need to identify relationships between variables when the relationship may hold only over a subset of the observations available. For example, expression of a particular gene may cause resistance to an anticancer drug, but only over certain types of cancer cell-lines. It may not be known in advance which types of cancer cell-lines (e.g., estrogen-regulated, newly differentiated, central nervous system) employ such a method of resistance. This situation differs from the usual setting in which partial correlations are estimated conditional on a known selection, such as the value of another variable. For any pair of variables of interest, the goal is to test if these are associated in some unspecified subpopulation that is represented by a subsample of the data we have available. Previous approaches rely heavily on bivariate normal assumptions, which are not easily adapted to non-linear association. We have tried several parametric and non-parametric approaches, and for both inferential and computational reasons have chosen to present a procedure based on a sequential development of Kendallʼs tau measure of monotone association. The sequence is achieved by reordering observations so that the sample tau coefficients {Tk} for the first k=2,...,n of the n observations form a monotone decreasing path, ending at Kendallʼs tau coefficient Tn. Boundaries are constructed so that 95% of the paths remain within the boundaries under the null hypothesis of independence. A boundary crossing at any point k is evidence of a\ stronger than expected association amongst a subpopulation represented by the k observations involved. The method is used to screen for association between gene expression and compound activity amongst types of cancer cell-lines in the NCI-60 database.
More generally, the method may be used to deconvolve a mixture of absolutely continuous bivariate distributions with the same margins but differing in strength of association. We prove that a particular method of reordering the observations is optimal against any other ordering for simultaneously identifying the highest τ association in subsets of size k (k=2,...,n). Furthermore, assuming a subpopulation of k, we present a way of quantifying how likely any observation is to be in that subpopulation. Extensions of this basic method to conditional testing, survival data, and time-series are described briefly.
This is joint work with Li Yu (MedImmune) and Paul Blower (Leadscope).
Bio:
Joe graduated from MIT with an SB in Mathematics, completed his Ph.D. at Stanford in Statistics, and did post-doctoral work at Carnegie-Mellon before taking a permanent position in the Statistics Department at the Ohio State University. He is a fellow of the American Statistical Association (ASA), and has won the ASA Presidential Award for his work on Statistics in Chemistry. A previous Fulbright fellow, he is currently Editor-in-Chief of Statistical Analysis and Data Mining and Founding Chair of the ASA Section on Statistical Learning and Data Mining.
You are cordially invited to attend a reception in the lounge next to Klaus 1324 before the seminar to chat informally with faculty and students. PIZZA will be provided.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>For more information please contact Dr. George Biros at biros@cc.gatech.edu
"Load-Balanced Bonded Force Calculations on Anton"
Abstract:
Molecular dynamics (MD) simulations of biological molecules involve the calculation of "bonded" force terms due to covalent bonds. On parallel computers, this calculation is normally not load balanced because the data partitioning required for load balance is in conflict with the partitioning needed for the much more expensive "nonbonded" (electrostatic and van der Waals) force calculation. On Anton, a specialized parallel machine for MD calculations, the situation has changed: the nonbonded component of the calculation has been dramatically accelerated and the remaining bonded component can often be a non-negligible determinant of overall performance. This talk describes the challenges in load balancing the bonded force calculations on Anton. In addition to the usual considerations of balancing load and minimizing communication across processors, we consider balancing the storage required per node, to allow larger chemical systems to be simulated. This interesting combinatorial problem arises because many bond terms share the same parameter data and this data does not need to be duplicated within a node. We also consider the hierarchical problem of partitioning the data among Anton nodes and its relation to partitioning the data within each node for computation by multiple cores. Additional resource limitations at this level also lead to interesting (and messy!) combinatorial problems for load balance.
]]>Phone:404-385-4785
lometa@cc.gatech.edu
]]>For more information please contact Dr. George Biros at biros@cc.gatech.edu
"Some variational approaches for multiphase segmentation"
Abstract:
Image segmentation has been widely studied, specially since Mumford-Shah functional was been proposed. Many theoretical works as well as numerous extensions have been studied rough out the years. This talk will focus on variational approaches for multi-phase segmentation. For the first model, we propose a model built upon the phase transition model of Modica and Mortola in material sciences and a properly synchronized fitting term that complements it. For the second model, we propose a variational functional for an unsupervised multiphase segmentation, by adding scale information of each phase. This model is able to deal with the instability issue associated with choosing the number of phases for multiphase segmentation. Finally, I will discuss some new results of scale segmentation which is an extension from the unsupervised model.
Bio:
http://people.math.gatech.edu/~kang/
You are cordially invited to attend a reception in the lounge next to Klaus 1324 before the seminar to chat informally with faculty and students. PIZZA will be provided.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Phone:404-385-4785
lometa@cc.gatech.edu
]]>For more information please contact Dr. David Bader
"Accurate Inference of Phylogenetic Relationships from Multi-locus Data"
Abstract:
Accurate inference of phylogenetic relationships of species, and understanding their relationships with gene trees are two central themes in molecular and evolutionary biology. Traditionally, a species tree is inferred by (1) sequencing a genomic region of interest from the group of species under study, (2) reconstructing its evolutionary history, and (3) declaring it to be the estimate of the species tree. However, recent analyses of increasingly available multi-locus data from various groups of organisms have demonstrated that different genomic regions may have evolutionary histories (called“gene trees”) that may disagree with each other, as well as with that of the species. This observation has called into question the suitability of the traditional approach to species tree inference. Further, when some, or all, of these disagreements are caused by reticulate evolutionary events, such as hybridization, then the phylogenetic relationship of the species is more appropriately modeled by a phylogenetic network than a tree. As a result, a new, post-genomic paradigm has emerged, in which multiple genomic regions are analyzed simultaneously, and their evolutionary histories are reconciled in order to infer the evolutionary history of the species, which may not necessarily be treelike. In this talk, I will describe our recent work on developing mathematical criteria and algorithmic techniques for analyzing incongruence among gene trees, and inferring phylogenetic relationships among species despite such incongruence. This includes work on lineage sorting, reticulate evolution, as well as simultaneous treatment of both. If time permits, I will describe our recent work on population genomic analysis of bacterial data,and the implications on the evolutionary forces shaping the genomic diversity in these populations.
Bio:
Luay Nakhleh is an Assistant Professor of Computer Science and Biochemistry and Cell Biology at Rice University, and an adjunct Assistant Professor of Systems Biology at MD Anderson Cancer Center. He received the B.Sc. degree from the Technion, Israel Institute of Technology, in 1996, the Master’s degree from Texas A&M University in 1998, and the PhD degree from the University of Texas at Austin in 2004⎯all three degrees in Computer Science. His research interests fall into the general areas of computational biology and bioinformatics; in particular, he works on computational phylogenomics and population genomics, and their connections with other fields in biology. Luay has published over 50 manuscripts on his work, supervised the dissertations of two recent PhD graduates, and currently supervises the dissertations of 6 PhD students. Luay has received several awards, including the Texas Excellence Teaching Award from UT Austin in 2001, the Outstanding Dissertation Award from UT Austin in 2005, the Roy E. Campbell Faculty Development Award from Rice University in 2006, the DOE Early Career Award in 2006, the NSF CAREER Award in 2009, the Phi Beta Kappa Teaching Prize in 2009, and an Alfred P. Sloan Foundation Fellowship in 2010.
]]>Phone:404-385-4785
lometa@cc.gatech.edu
]]>For more information please contact Dr. David Bader
"Challenges in Combinatorial Scientific Computing"
Abstract:
Computation on large combinatorial structures -- graphs, strings, partial orders, etc. -- has become fundamental in many areas of data analysis and scientific modeling. The field of high-performance combinatorial computing, however, is in its infancy. By way of contrast, in numerical supercomputing we possess standard algorithmic primitives, high-performance software libraries, powerful rapid-prototyping tools, and a deep understanding of effective mappings of problems to high-performance computer architectures.
This talk will describe a number of challenges for the field of combinatorial scientific computing, in algorithms, tools, architectures, and mathematics. I will draw examples from various applications, and I will highlight our group's work on algebraic primitives for high-performance computation on large graphs and networks.
Bio:
http://www.cs.ucsb.edu/~gilbert/
You are cordially invited to attend a reception in the lounge next to Klaus 1324 before the seminar to chat informally with faculty and students. PIZZA will be provided.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>404-385-4785
]]>For more information please contact Dr. George Biros at gbiros@cc.gatech.edu
~~~~~~~~~~~~~~~~~~~~~~
Title:
Efficient High-Order Discontinuous Galerkin Methods for Fluid Flow Simulations
Abstract:
The inadequacy of current production-level computational fluid dynamics codes in delivering sufficient accuracy in numerical flow simulations, as well as in resolving a wide range of turbulence scales for reliable large eddy simulations have been widely realized over the past decade. Since these codes are typically based on low-order finite volume methods, high-order methods such as discontinuous Galerkin (DG) methods have been advocated as alternative discretization techniques. DG methods are weighted residual methods with discontinuous approximate solution spaces typically consisting of polynomials defined on each element of the geometry triangulation. The inter-element connectivities are enforced through a proper definition of numerical fluxes along the shared boundaries between elements. Like more traditional finite element methods (continuous Galerkin methods), DG methods are well-suited for simulations in complex geometries. Moreover, they offer advantages in capturing features of convection-dominated flows, facilitating hp-adaptivity, ease of parallelization, and in the effectiveness of block diagonal iterative solvers. For these advantages to be realized in industrial simulations, efficient solution strategies should be developed for systems arising from high-order DG discretizations. Addressing this issue in the context of solving both incompressible and compressible Navier-Stokes equations is the focus of this talk.
For the unsteady incompressible Navier-Stokes equations, we present an efficient parallel solver based on high-order DG methods using triangular and tetrahedral meshes in two and three space dimensions, respectively. In the context of semi-explicit temporal discretization, we present an algorithm with compact stencil sizes for all discrete equations yielding minimum computation and communication costs.
For the compressible Navier-Stokes equations, we present multigrid algorithms for systems arising from high-order DG discretizations on unstructured meshes. The algorithms are based on coupling both functional and geometric multigrid methods which are used in non-linear or linear forms, and either directly as solvers or as preconditioners to a Newton-Krylov method.
We identify two areas that need further explorations, namely developing new high-order DG algorithms for high-speed flows involving strong shock discontinuities, and methods for efficient simulation of flow-structure interactions. Finally, we describe the motivation behind our current research effort on devising hybrid Fourier continuation-WENO finite difference solvers for simulating high-speed multi-material compressible flows such as Richtmyer-Meshkov instabilities. This is an ongoing collaboration with Prof. Oscar Bruno's research group at the California Institute of Technology.
Bio:
http://www.dam.brown.edu/people/shahbazi/
]]>404.385.4785
]]>For more information please contact Dr. Rich Vuduc at richie@cc.gatech.edu
"Ocelot: An Infrastructure for Analysis and Workload Characterization of Accelerated Applications"
Abstract:
The rapid emergence of GPGPU architectures and their integration into commodity systems has produced significant productivity and software development challenges. These challenges are accompanied by a dearth of productivity tools for application development and analysis. In particular, the migration of existing compute intensive applications to these new CPU-GPGPU platforms suffers from a lack of productivity tools that can provide insights into candidate code segments that can benefit from acceleration, an assessment of most appropriate target platform, debugging, and support for subsequent performance tuning.
This talk will describe the Ocelot infrastructure being developed to address this lack of productivity tools for accelerated computing. Ocelot is emulation, compilation, and workload characterization environment targeted for CPUs accelerated with NVIDIA GPUs. The talk will describe our vision for Ocelot and progress date. Using Ocelot, CUDA kernels are transparently executed on a target GPU device, emulated in software in the host, or translated to be executed natively on a multithreaded multicore host (e.g., x86). Ocelot is being integrated with the Harmony runtime being developed in our group to enable performance portability and binary portability across platform sizes, e.g., different numbers and types of NVIDIA accelerators. Examples are presented of the analysis of emerging benchmarks in the research community such as Parboil, Rodinia, and the CUDA SDK.
Bio:
Sudhakar Yalamanchili earned his PhD degree in Electrical and Computer Engineering in 1984 from the University of Texas at Austin after which he joined Honeywell’s Systems and Research Center in Minneapolis where he worked as a Senior, and then Principal Research Scientist from 1984 to 1989. In both capacities, he served as the Principal Investigator for projects in the design and analysis of multiprocessor architectures for embedded applications. While at Honeywell, Dr. Yalamanchili also served as an Adjunct Faculty and taught in the Department of Electrical Engineering at the University of Minnesota. He joined the ECE faculty at Georgia Tech in 1989 where he is now a Joseph M. Pettit Professor of Computer Engineering. He is the author of VHDL Starters Guide, 2nd edition, Prentice Hall 2004, VHDL: From Simulation to Synthesis, Prentice Hall, 2000, and co-author with J. Duato and L. Ni, of Interconnection Networks: An Engineering Approach, Morgan Kaufman, 2003.
Dr. Yalamanchili is a Senior Member of the IEEE. He has served as a Distinguished Visitor of the IEEE, and on the editorial boards of the IEEE Transactions on Parallel and Distributed Processing and IEEE Transactions on Computers. He contributes professionally through service on conference and workshop program committees in the area of high performance computing, computer architecture, and interconnection networks.
~~~~~~~~~~~~~~~~
You are cordially invited to attend a reception in the lounge next to Klaus 1324 before the seminar to chat informally with faculty and students. PIZZA will be provided.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Phone: 404-385-4785
Email: lometa@cc.gatech.edu
]]>For more information please contact Dr. George Biros at gbrios@cc.gatech.edu
"Accelerating Parallel Analysis of Scientific Simulation Data via Zazen"
Abstract:
As a new generation of parallel supercomputers enables researchers to conduct scientific simulations of unprecedented scale and resolution, terabyte-scale simulation output has become increasingly commonplace. Analysis of such massive data sets is typically I/O-bound: many parallel analysis programs spend most of their execution time reading data from disk rather than performing useful computation. To overcome this I/O bottleneck, we have developed a new data access method. Our main idea is to cache a copy of simulation output files on the local disks of an analysis cluster’s compute nodes, and to use a novel task-assignment protocol to co-locate data access with computation. We have implemented our methodology in a parallel disk cache system called Zazen. By avoiding the overhead associated with querying metadata servers and by reading data in parallel from local disks, Zazen is able to deliver a sustained read bandwidth of over 20 gigabytes per second on a commodity Linux cluster with 100 nodes, approaching the optimal aggregated I/O bandwidth attainable on these nodes. Compared with conventional NFS, PVFS2, and Hadoop/HDFS, respectively, Zazen is 75, 18, and 6 times faster for accessing large (1-GB) files, and 25, 13, and 85 times faster for accessing small (2-MB) files. We have deployed Zazen in conjunction with Anton—a special-purpose supercomputer that dramatically accelerates molecular dynamics (MD) simulations—and have been able to accelerate the parallel analysis of terabyte-scale MD trajectories by about an order of magnitude.
Bio:
Tiankai Tu is a computer scientist at D. E. Shaw Research, where he architects and implements parallel software systems for analyzing very long molecular dynamics trajectories. He is also a visiting scientist at the University of Texas at Austin, developing parallel adaptive mesh refinement (AMR) algorithms for simulating global mantle convection on petascale computers. Tiankai earned a Ph.D. in Computer Science from Carnegie Mellon University, where he developed computational database systems and parallel algorithms for simulating earthquake ground motion on terascale systems.
He received the Gordon Bell Award for Special Achievement in 2003, the SC06 HPC Analytics Challenge Award in 2006, the TeraGrid Capability Computing Challenge Award in 2008, and the SC09 Best Poster Award in 2009. He was also a finalist for the SC06 Best Student Paper Award, the SC08 Best Technical Paper Award, and the 2008 Gordon Bell Award for Special Achievement.
~~~~~~~~~~~~~~~~
You are cordially invited to attend a reception in the lounge next to Klaus 1324 before the seminar to chat informally with faculty and students. Refreshments will be provided.
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Phone: 404-385-4785
Email: lometa@cc.gatech.edu
]]>For more information please contact Dr. David Bader
~~~~~~~~~~~~~~~~~~~~~~
"Discovery of Mechanisms from Mathematical Modeling of DNA Microarray Data: Computational Prediction and Experimental Verification"
Abstract:
Future discovery and control in biology and medicine will come from the mathematical modeling of large-scale molecular biological data,such as DNA microarray data, just as Kepler discovered the laws of planetary motion by using mathematics to describe trends in astronomical data [1]. In this talk, I will demonstrate that mathematical modeling of DNA microarray data can be used to correctly predict previously unknown mechanisms that govern the activities of DNA and RNA.
First, I will describe the computational prediction of a mechanism of regulation, by developing generalizations of the matrix and tensor computations that underlie theoretical physics and using them to uncover a genome-wide pattern of correlation between DNA replication initiation and RNA expression during the cell cycle [2,3].
Second, I will describe the recent experimental verification of this computational prediction, by analyzing global expression in synchronized cultures of yeast under conditions that prevent DNA replication initiation without delaying cell cycle progression [4].
Third, I will describe the use of the singular value decomposition to uncover "asymmetric Hermite functions," a generalization of the eigenfunctions of the quantum harmonic oscillator, in genome-wide mRNA lengths distribution data [5]. These patterns might be explained by a previously undiscovered asymmetry in RNA gel electrophoresis band
broadening and hint at two competing evolutionary forces that determine the lengths of gene transcripts.
Finally, I will describe ongoing work in the development of tensor algebra algorithms (as well as viusal correlation tools), the integrative and comparative modeling of DNA microarray data (as well as rRNA sequence data), and the discovery of mechanisms that regulate cell division, cancer and evolution.
Bio:
http://www.bme.utexas.edu/research/orly/alter/biography.html
Phone: 404-385-4785
Email: lometa@cc.gatech.edu
]]>For more information please contact Dr. Haesun Park at hpark@cc.gatech.edu
"Developing a Visual Analytics Approach to Analytic Problem-Solving"
Abstract:
Visual analytics attacks large scale, complex, often incomplete data. What has become clear is that not only are the data complex, but the application problems are complex as well, often involving reasoning, insight discovery, hypothesis-building, evidence gathering, and actionable problem-solving. A major challenge of visual analytics is to offer a new integration of interactive visualization and analytic methods so that large scale, complex data and the complex problems from which they arise can be attacked together successfully. In this talk, I will present our approach to this challenge. I will illustrate this approach with some applications.
Bio:
William Ribarsky is the Bank of America Endowed Chair in Information Technology at UNC Charlotte and the founding director of the Charlotte Visualization Center. He is currently Chair of the Computer Science Department. Dr. Ribarsky is Principal Investigator for the DHS SouthEast Regional Visualization and Analytics Center. He received a Ph.D. in physics from the University of Cincinnati. His research interests include visual analytics; 3D multimodal interaction; bioinformatics visualization; virtual environments; visual reasoning; and interactive visualization of large-scale information spaces. Dr. Ribarsky is the former Chair and a current Director of the IEEE Visualization and Graphics Technical Committee. He is also a member of the Steering Committees for the IEEE Visualization Conference and the IEEE Virtual Reality Conference, the leading international conferences in their fields. He was an Associate Editor of IEEE Transactions on Visualization and Computer Graphics and is currently an Editorial Board member for IEEE Computer Graphics & Applications. Dr. Ribarsky co-founded the Eurographics/IEEE visualization conference series (now called EG/IEEE EuroVis) and led the effort to establish the current Virtual Reality Conference series. For the above efforts on behalf of IEEE, Dr. Ribarsky won the IEEE Meritorious Service Award in 2004. In 2007, he was general co-chair of the IEEE Visual Analytics Science and Technology (VAST) Symposium.
Dr. Ribarsky has published over 130 scholarly papers, book chapters, and books. He has received competitive research grants and contracts from NSF, ARL, ARO, DHS, ONR, EPA, AFOSR, DARPA, NASA, NIMA, US DOT, National Institute of Justice, and several companies.
You are cordially invited to attend a reception in the lounge next to Klaus 1324 before the seminar to chat informally with faculty and students. PIZZA will be provided.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>404.385.4785
]]>"Generalized Fast Multipole Method"
Abstract:
I will present a novel FMM technique applicable to both oscillatory and non-oscillatory
kernels. The method is very flexible and accurate. It is more general than existing FMMs and can often compete with nearly optimal schemes for specific kernels like 1/r. I will present the progress of the research with some preliminary numerical results.
Bio:
You are cordially invited to attend a reception preceding the seminar to chat informally with faculty and students. Reception will take place outside Klaus 1324 between 1:30 - 2:00.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Phone: 404-385-4785
Email: lometa@cc.gatech.edu
]]>Professor, Mathematics and Computer Science Department, Emory University
"Efficient Iterative Methods for Large Scale Inverse Problems"
Abstract:
Ill-posed problems arise in many image processing applications, including microscopy, medicine and astronomy. Iterative methods are typically recommended for these large scale problems, but they can be difficult to use in practice. In this talk we describe a hybrid approach that combines the Golub-Kahan bidiagonalization algorithm with Tikhonov regularization and a weighted generalized cross validation scheme. We also show how this method can be used as an effective scheme for choosing regularization parameters for certain nonlinear inverse problem. Applications from image processing illustrate the effectiveness of the resulting numerical schemes.
This is joint work with Julianne Chung, University of Maryland.
Bio:
James Nagy is a Professor of Mathematics and Computer Science at Emory University. He received his Ph.D. in Applied Mathematics from North Carolina State University in 1991. Before joining Emory University in 1999 he had postdoctoral research fellowships with the IMA at the University of Minnesota, with the NSF at the University of Maryland, and was on the faculty at Southern Methodist University. He is on the editorial boards of SIAM Journal on Scientific Computing (SISC), SIAM Journal on Matrix Analysis and Applications (SIMAX), and the SIAM Book Series "Fundamentals of Algorithms". His research interests include numerical linear algebra, structured matrix computations, and numerical solution of inverse problems in image processing. Among his many publications includes the recent SIAM book, "Image Deblurring: Matrices, Spectra and Filtering", which is co-authored with Per Christian Hansen and Dianne O'Leary. Visit Dr. Nagy's website.
Please join us for a reception preceding the seminar outside Klaus 1324, beginning at 1:30pm.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Lometa Mitchell
Phone:404-385-4785
Email:lometa@cc.gatech.edu
"Model-Based Data Mining for Functional Data Under Spatial Interdependence"
Abstract:
In this seminar, I will present data mining methods for discovering and summarizing patterns in functional data observed under spatial interdependence. The field of functional data analysis has already provided a series of competitive data mining approaches, but they are generally limited to the assumption of independence between the random functions. This assumption is rather restrictive in many research applications.
In the first part of this seminar, I will introduce a model-based method for clustering random functions which are spatially interdependent. The time functions are decomposed into spatial global and time-dependent cluster effects using a semi-parametric model. We assume that the clustering membership is a realization from a Markov random field. In the case study presented in this paper, we focus on obtaining temporal cluster trends for racial-ethnic diversity for five southeast states in the US.
In the second part of this seminar, I will introduce a computational efficient and theoretically-founded cross-correlation analysis. Under the proposed semi-parametric model, we show that the cross-correlation estimators are asymptotically unbiased under the conditions that the sample size is large and the intrinsic dimensionality of the functional processes is much smaller than the sample size. We illustrate this correlation analysis within a demographic study, in which we analyze the association between per capita income and racial-ethnic diversity.
This is joint work with Huijing Jiang, PhD student in ISyE, Georgia Institute of Technology
Bio:
Nicoleta Serban is an assistant professor in ISyE. She received her B.S. in Mathematics and an M.S. in Theoretical Statistics and Stochastic Processes from the University of Bucharest. She went on to earn her Ph.D. in Statistics at Carnegie Mellon University. Before joining Georgia Tech, Dr. Serban's research focused on nonparametric statistical methods motivated by recent applications from proteomics and genomics. Dr Serban's current research focusses on multiple functional estimation and clustering with applications to industrial performance, service site location, socio-economics and NMR biomolecular studies. Visit Dr. Serban's website.
Please join us for a reception preceding the seminar outside Klaus 1324, beginning at 1:30pm.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
Lometa Mitchell
Phone:404-385-4785
Email:lometa@cc.gatech.edu
For more information please contact Dr. George Biros at gbiros@cc.gatech.edu
"Nonlinear Models Motivated by Manifold-Based Learning"
Abstract:
In statistical modeling, a well-adopted framework is as follows: consider y = f(X), where response y is univariate, predictor X can be multivariate, and function f is what we try to identify. Such a formulation covers nearly all statistical models that we have encountered: e.g., linear (regression) model or parametric models with g being linear or belonging to a specific parametric family; nonlinear models such as kernel methods, splines, local polynomials, etc., when g has a particular form. We consider a variation of the above by reconsidering the penalized estimation framework: finding f which optimize a function having form G(f) + R(f), where G(f) is called a goodness of fit measure and R(f) measures the regularity of the functional solution f. It is known that under certain circumstances, the solutions to the above has a close-form solutions: e.g., when G(f) is the residuals sum of squares and R(f) is the integrated square of the second derivative of f, the minimizer f is the finite-dimensional cubic smoothing splines. It is also known that the above does not have a direct generalization to high-dimensional X. On the other hand, the date with high-dimensional X is frequently encountered in practice. We introduce an alternative to R(f), which leads to close-form solution and easy numerical implementation. Experiments with both synthetic and real data demonstrate the superiority of our method. The resulting model is nonlinear, not covered by any existing family of models, and requires very little assumption on the underlying f. Potential theoretical results will be discussed. The newly introduced R(f) is motivated by the work of hessian eigenmaps -- a manifold-based dimension reduction algorithm.
Bio:
Xiaoming Huo received the B.S. degree in mathematics from the University of Science and Technology, China, in 1993, and the M.S. degree in electrical engineering and the Ph.D. degree in statistics from Stanford University, Stanford, CA, in 1997 and 1999, respectively. Since August 2006, he has been an Associate Professor with the School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta. He represented China in the 30th International Mathematical Olympiad (IMO), which was held in Braunschweig, Germany, in 1989, and received a golden prize.
His research interests include statistics and multiscale methodology. He has made numerous contributions on topics such as sparse representation, wavelets, and statistical problems in detectability. His papers appeared in top journals, and some of them are highly cited. See his publication list for details.
Please join us for a reception preceding the seminar outside Klaus 1324, beginning at 1:30pm.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Lometa Mitchell
Phone:404-385-4785
Email:lometa@cc.gatech.edu
]]>For more information please contact Dr. Richard Vuduc at richie@cc.gatech.edu
"Parallel Thinking"
Abstract:
With the advent of manycore computers on every desktop and the recent access by just about anyone to massively parallel data centers, parallelism is becoming pervasive. This trend is likely to eventually lead to a scenario in which parallel programming will become predominant and sequential programming will be a special case. Are we ready for this change? Short-term solutions based just on add-ons to sequential languages are unlikely to be sufficient in the long term. Instead the change will likely require a more fundamental rethinking that permeates the programmers' methodologies from early stages of algorithm and system design. This will require developing a form of "parallel thinking."
Perhaps the biggest barrier to the widespread effective use of parallelism is educating people on how to think parallel. Many if not most computer science classes, however, remain case studies in how to push students into thinking sequentially. This talk will address how parallelism could be taught right from the start, and if presented at the right level of abstraction could be no harder than teaching sequential programming.
Bio:
Guy Blelloch is a Professor of Computer Science. His research interests are in programming languages and algorithms and how they interact with an emphasis on parallel computation. He worked on one of the early Parallel Machines, the Thinking Machines Connection Machine, where he developed several of the parallel primitives for the machine. At Carnegie Mellon Blelloch designed and implemented the parallel programming language NESL, a language designed for easily expressing and analyzing parallel algorithms. Other more recent work on parallelism has addressed issues in scheduling, algorithm design, cache efficiency, garbage collection, and synchronization primitives. Visit Dr. Blelloch's website.
~~~~~~~~~~~~~~~~~~~~~~~
Please join us for a reception preceding the seminar outside Klaus 1324, beginning at 1:30pm.
To receive future announcements, please sign up to the cse-seminar email list: https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Lometa Mitchell
Phone:404-385-4785
Email:lometa@cc.gatech.edu
For more information please contact Dr. Guy Lebanon at lebanon@cc.gatech.edu
"Robust algorithms for information retrieval: effective tradeoffs between risk and reward"
Abstract:
Current search engines must typically act under great uncertainty, attempting to interpret the intent of few keywords from the user to match against billions of documents to find a few relevant results. A typical Web search involves a series of operations based on the user's query, from automatic spelling correction and identifying common word variants to the actual document ranking. Such operations tend to have a risk-reward tradeoff depending on how they "bet" on different solution hypotheses. In this talk I discuss new theoretical models, algorithms, and evaluation methods for estimating and accounting for uncertainty in retrieval algorithms to achieve effective risk-reward tradeoffs.
A prime example of a high-risk, high-reward operation is automatic query reformulation that adds related terms to a query - a process known as query expansion. Query expansion can significantly improve ranking quality on average, but even state-of-the-art methods are highly unreliable and can significantly hurt result quality for some queries, which is one reason for their limited deployment in real-world scenarios. I discuss how casting query expansion as a
constrained optimization problem over a word graph provides a selective, highly effective modeling framework that reduces the number and magnitude of expansion failures with no loss in the strong average-case gain of the underlying expansion algorithm. I also discuss applications of such optimization frameworks to other problems in information retrieval, such as providing diversity in ranking.
Bio:
Kevyn Collins-Thompson is a Researcher in the Context, Learning and User Experience for Search (CLUES) group at Microsoft Research. He completed his PhD (2008) in Computer Science at the Language Technologies Institute, Carnegie Mellon University, where his advisor was Jamie Callan. His research focuses on theoretical models, algorithms, and evaluation methods for effective, reliable information retrieval. Other research interests include user models and personalization, modeling evolutional dynamics of text, educational applications of search technology and machine learning, and understanding how the brain acquires language skills. Kevyn also has more than ten years of industry experience as a software engineer and project manager, responsible for shipping advanced features in Office, Windows, Tablet PC, Encarta, and many other products. Visit his website here.
~~~~~~~~~~~~~~
Please join us for a reception preceding the seminar outside Klaus 1324, beginning at 1:30 pm
To receive future announcements, please sign up to the cse-seminar email list:
https://mailman.cc.gatech.edu/mailman/listinfo/cse-seminar
]]>Lometa Mitchell
Phone:404-385-4785
Email:lometa@cc.gatech.edu
"The LP2 library: an easy way to multi-thread programs dealing with unstructured meshes"
Abstract:
"As of today, programers willing to multi-thread their code are provided with many choices like OpenMP, UPC or various compilers featuring automated parallelization. Unfortunately, they provide no easy way to handle loops that perform indirect memory accesses, which are common ground when dealing with unstructured meshes. The LP2 allows for parallelization of such loops with the help of dynamic scheduling and space filling curves.If I have enough time, I will also discuss HEXOTIC, an automated hexahedral mesh generator combining octree, dual-mesh and buffer-layers. Even though many methods have been suggested to meet the challenge of all-hexahedral meshing, octree-based methods remain the most efficient from an engineering point of view. As of today, its speed and robustness are still unmatched. This paper presents advances made in the Hexotic project, especially in terms of automation, sharp angles meshing and adaptation."
Bio:
"I have been working on automated mesh generation for twelve years in the Gamma team at INRIA / FRANCE. My main research subjects are, hexahedral meshing, adaptive meshing and surface meshing. I am also working on multi-threading as a side project"
Pizza will be served
]]>Phone: 404-385-4785
Email: lometa@cc.gatech.edu
]]>a talk by
Friday, September 2 at 11 a.m.
SpaceX's reusable rocket program aims to reduce the cost of space travel by making rockets that can land, refuel and refly, instead of being thrown away after every flight. Precise landing of a rocket is a unique problem, which has been likened to balancing a rubber broomstick on your hand in a windstorm. Rockets do not have wings (unlike airplanes) and they cannot rely on a high ballistic coefficient to fly in a straight line (unlike missiles). In the past year, SpaceX has successfully landed five rockets, two of which were on dry land at Cape Canaveral, and three of which were on a floating platform in the Atlantic. This talk will discuss the challenges involved, how these challenges were overcome, and next steps towards rapid reusability.
]]>