{"671677":{"#nid":"671677","#data":{"type":"event","title":"ISyE Seminar - Sen Na","body":[{"value":"\u003Ch3\u003ETitle:\u003C\/h3\u003E\r\n\r\n\u003Cp\u003EPracticality meets Optimality: Real-Time Statistical Inference under Complex Constraints\u003C\/p\u003E\r\n\r\n\u003Ch3\u003EAbstract:\u003C\/h3\u003E\r\n\r\n\u003Cp\u003EConstrained estimation problems are prevalent in statistics, machine learning, and engineering. These problems\u0026nbsp;encompass constrained generalized linear models, constrained deep neural networks, physics-inspired machine\u0026nbsp;learning, algorithmic fairness, and optimal control. However, existing estimation methods under hard constraints\u0026nbsp;rely on either projection or regularization, which may theoretically exhibit optimal efficiency but are impractical\u0026nbsp;or unreasonably fail in reality. This talk aims to bridge the significant gap between practice and theory\u0026nbsp;for constrained estimation problems.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EI will begin by introducing the critical methodology used to bridge the gap, called Stochastic Sequential Quadratic\u0026nbsp;Programming. We will see that SQP methods serve as the workhorse for modern scientific machine learning problems\u0026nbsp;and can resolve the failure modes of prevalent regularization-based methods. I will demonstrate how to make\u0026nbsp;SQP adaptive and scalable using various modern techniques, such as stochastic line search, trust region, and dimension\u0026nbsp;reduction.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EAdditionally, I will show how to further enhance SQP to handle inequality constraints online.\u003Cbr \/\u003E\r\nFollowing the methodology, I will present some selective theories, emphasizing the consistency and efficiency\u003Cbr \/\u003E\r\nof the SQP methods. Specifically, I will show that online SQP iterates asymptotically exhibit normal behavior\u003Cbr \/\u003E\r\nwith a mean of zero and optimal covariance in the H\u00e1jek and Le Cam sense. Significantly, the covariance does\u003Cbr \/\u003E\r\nnot deteriorate even when we apply modern techniques driven by practical concerns. The talk concludes with\u003Cbr \/\u003E\r\nexperiments on both synthetic and real datasets.\u003C\/p\u003E\r\n\r\n\u003Ch3\u003EBio:\u003C\/h3\u003E\r\n\r\n\u003Cp\u003ESen Na is currently a postdoctoral researcher in the Department of Statistics and the International Computer\u003Cbr \/\u003E\r\nScience Institute at UC Berkeley. He received a Ph.D. degree in statistics from the University of Chicago.\u003Cbr \/\u003E\r\nSen Na\u2019s primary research interests lie in the mathematical foundations of data science, encompassing high dimensional\u0026nbsp;statistics, computational statistics, sequential decision-making, and large-scale and stochastic\u003Cbr \/\u003E\r\nnonlinear optimization. Additionally, he is passionate about various applications of machine learning methods in\u0026nbsp;scientific fields such as biology, neuroscience, physics, and engineering. Sen Na\u2019s research has been recognized\u0026nbsp;by the prestigious Harper Dissertation Fellowship from UChicago, and he has been selected as one of the\u0026nbsp;Young Researchers in ORIE by Cornell University.\u003C\/p\u003E\r\n","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Ch3\u003EAbstract:\u003C\/h3\u003E\r\n\r\n\u003Cp\u003EConstrained estimation problems are prevalent in statistics, machine learning, and engineering. These problems\u003Cbr \/\u003E\r\nencompass constrained generalized linear models, constrained deep neural networks, physics-inspired machine\u003Cbr \/\u003E\r\nlearning, algorithmic fairness, and optimal control. However, existing estimation methods under hard constraints\u003Cbr \/\u003E\r\nrely on either projection or regularization, which may theoretically exhibit optimal efficiency but are impractical\u003Cbr \/\u003E\r\nor unreasonably fail in reality. This talk aims to bridge the significant gap between practice and theory\u003Cbr \/\u003E\r\nfor constrained estimation problems.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EI will begin by introducing the critical methodology used to bridge the gap, called Stochastic Sequential Quadratic\u003Cbr \/\u003E\r\nProgramming. We will see that SQP methods serve as the workhorse for modern scientific machine learning problems\u0026nbsp;and can resolve the failure modes of prevalent regularization-based methods. I will demonstrate how to make\u0026nbsp;SQP adaptive and scalable using various modern techniques, such as stochastic line search, trust region, and dimension\u0026nbsp;reduction. Additionally, I will show how to further enhance SQP to handle inequality constraints online.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EFollowing the methodology, I will present some selective theories, emphasizing the consistency and efficiency\u003Cbr \/\u003E\r\nof the SQP methods. Specifically, I will show that online SQP iterates asymptotically exhibit normal behavior\u003Cbr \/\u003E\r\nwith a mean of zero and optimal covariance in the H\u00e1jek and Le Cam sense. Significantly, the covariance does\u003Cbr \/\u003E\r\nnot deteriorate even when we apply modern techniques driven by practical concerns. The talk concludes with\u003Cbr \/\u003E\r\nexperiments on both synthetic and real datasets.\u003Cbr \/\u003E\r\n\u0026nbsp;\u003C\/p\u003E\r\n","format":"limited_html"}],"field_summary_sentence":[{"value":"Practicality meets Optimality: Real-Time Statistical Inference under Complex Constraints"}],"uid":"34977","created_gmt":"2023-12-21 14:32:47","changed_gmt":"2023-12-21 14:35:28","author":"Julie Smith","boilerplate_text":"","field_publication":"","field_article_url":"","field_event_time":{"event_time_start":"2024-01-16T11:00:00-05:00","event_time_end":"2024-01-16T12:00:00-05:00","event_time_end_last":"2024-01-16T12:00:00-05:00","gmt_time_start":"2024-01-16 16:00:00","gmt_time_end":"2024-01-16 17:00:00","gmt_time_end_last":"2024-01-16 17:00:00","rrule":null,"timezone":"America\/New_York"},"location":"ISyE Groseclose 402","extras":[],"groups":[{"id":"1242","name":"School of Industrial and Systems Engineering (ISYE)"}],"categories":[],"keywords":[],"core_research_areas":[],"news_room_topics":[],"event_categories":[{"id":"1795","name":"Seminar\/Lecture\/Colloquium"}],"invited_audience":[{"id":"78761","name":"Faculty\/Staff"},{"id":"177814","name":"Postdoc"},{"id":"174045","name":"Graduate students"},{"id":"78751","name":"Undergraduate students"}],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[],"email":[],"slides":[],"orientation":[],"userdata":""}}}