{"610202":{"#nid":"610202","#data":{"type":"event","title":"ML@GT Talk \u2014 Bryan McCann, Salesforce","body":[{"value":"\u003Cp\u003EThe\u0026nbsp;\u003Ca href=\u0022http:\/\/ml.gatech.edu\/\u0022 target=\u0022_blank\u0022\u003EMachine Learning Center at Georgia Tech (ML@GT)\u003C\/a\u003E\u0026nbsp;is excited to welcome Bryan McCann from Salesforce to campus for a ML@GT\u0026nbsp;Talk.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EFor scheduling information, contact Mark Riedl at\u0026nbsp;riedl@cc.gatech.edu\u003Cbr \/\u003E\r\n\u003Cbr \/\u003E\r\n\u003Cstrong\u003EPlease \u003Ca href=\u0022https:\/\/docs.google.com\/forms\/d\/e\/1FAIpQLSeE3DGOwtEmXqJinD1vk958AGIAHHHv3YCv5RfI4DsKSyRgWA\/viewform?usp=sf_link\u0022\u003ERSVP\u003C\/a\u003E by Tuesday, August 27th.\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003ETitle\u003C\/strong\u003E\u003Cbr \/\u003E\r\nThe Natural Language Decathlon: Multitask Learning as Question Answering\u003Cbr \/\u003E\r\n\u003Cbr \/\u003E\r\n\u003Cstrong\u003EAbstract\u003C\/strong\u003E\u003Cbr \/\u003E\r\nDeep learning has improved performance on many natural language processing (NLP)\u0026nbsp;tasks individually. However, general NLP models cannot emerge within a paradigm that\u0026nbsp;focuses on the particularities of a single metric, dataset, and task.\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWe introduce the Natural Language Decathlon (decaNLP), a challenge that spans ten\u0026nbsp;tasks:\u003Cbr \/\u003E\r\nquestion answering, machine translation, summarization, natural language inference,\u0026nbsp;sentiment analysis, semantic role labeling, zero-shot relation extraction, goal-oriented\u0026nbsp;dialogue, semantic parsing, and commonsense pronoun resolution. We cast all tasks as question answering over a context. Furthermore, we present a new\u0026nbsp;Multitask Question Answering Network (MQAN) jointly learns all tasks in decaNLP without\u0026nbsp;any task-specific modules or parameters in the multitask setting. MQAN shows\u0026nbsp;improvements in transfer learning for machine translation and named entity recognition,\u0026nbsp;domain adaptation for sentiment analysis and natural language inference, and zero-shot\u0026nbsp;capabilities for text classification. We demonstrate that the MQAN\u0026#39;s multi-pointer-generator decoder is key to this success and performance further improves with an anti-curriculum training strategy.\u003Cbr \/\u003E\r\nThough designed for decaNLP, MQAN also achieves state of the art results on the\u0026nbsp;WikiSQL semantic parsing task in the single-task setting.\u003C\/p\u003E\r\n\r\n\u003Cp\u003EWe release code for procuring and processing data, training and evaluating models, and\u0026nbsp;reproducing all experiments for decaNLP.\u0026nbsp;\u003Cbr \/\u003E\r\n\u003Cbr \/\u003E\r\n\u003Cstrong\u003EBio\u003C\/strong\u003E\u003Cbr \/\u003E\r\n\u003Cbr \/\u003E\r\nBryan McCann is a Senior Research Scientist at Salesforce. He focuses on transfer learning and multitask\u0026nbsp;learning for natural language processing. Most recently, Bryan proposed the Natural Language Decathlon\u0026nbsp;(decaNLP) and a Multitask Question Answering Network to tackle all ten tasks in decaNLP. Before decaNLP,\u0026nbsp;he showed that the intermediate representations, or context vectors (CoVe), of machine translation systems\u0026nbsp;carry information that aids learning in question answering and text classification systems.\u003Cbr \/\u003E\r\nPrior to working at Salesforce, Bryan studied at Stanford University, where he completed a B.S and M.S in\u0026nbsp;Computer Science as well as a B.A in Philosophy.\u0026nbsp;\u003C\/p\u003E\r\n","summary":null,"format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003EThe\u0026nbsp;\u003Ca href=\u0022http:\/\/ml.gatech.edu\/\u0022 target=\u0022_blank\u0022\u003EMachine Learning Center at Georgia Tech (ML@GT)\u003C\/a\u003E\u0026nbsp;is excited to welcome Bryan McCann from Salesforce to campus for a ML@GT\u0026nbsp;Talk.\u003C\/p\u003E\r\n","format":"limited_html"}],"field_summary_sentence":[{"value":"\u0022The Natural Language Decathlon: Multitask Learning as Question Answering\u0022"}],"uid":"34773","created_gmt":"2018-08-22 18:31:11","changed_gmt":"2018-08-31 18:11:59","author":"ablinder6","boilerplate_text":"","field_publication":"","field_article_url":"","field_event_time":{"event_time_start":"2018-08-28T13:15:00-04:00","event_time_end":"2018-08-28T14:15:00-04:00","event_time_end_last":"2018-08-28T14:15:00-04:00","gmt_time_start":"2018-08-28 17:15:00","gmt_time_end":"2018-08-28 18:15:00","gmt_time_end_last":"2018-08-28 18:15:00","rrule":null,"timezone":"America\/New_York"},"extras":[],"hg_media":{"610198":{"id":"610198","type":"image","title":"Bryan McCann","body":null,"created":"1534962379","gmt_created":"2018-08-22 18:26:19","changed":"1534962379","gmt_changed":"2018-08-22 18:26:19","alt":"","file":{"fid":"232359","name":"Unknown-2.jpeg","image_path":"\/sites\/default\/files\/images\/Unknown-2_0.jpeg","image_full_path":"http:\/\/www.tlwarc.hg.gatech.edu\/\/sites\/default\/files\/images\/Unknown-2_0.jpeg","mime":"image\/jpeg","size":24629,"path_740":"http:\/\/www.tlwarc.hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/images\/Unknown-2_0.jpeg?itok=BJQNu0bg"}}},"media_ids":["610198"],"groups":[{"id":"47223","name":"College of Computing"},{"id":"576481","name":"ML@GT"}],"categories":[],"keywords":[{"id":"1808","name":"graduate students"}],"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":"78771","name":"Public"},{"id":"174045","name":"Graduate students"},{"id":"78751","name":"Undergraduate students"}],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003EAllie McFadden\u003C\/p\u003E\r\n\r\n\u003Cp\u003ECommunications Officer\u003C\/p\u003E\r\n\r\n\u003Cp\u003Eallie.mcfadden@cc.gatech.edu\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n","format":"limited_html"}],"email":[],"slides":[],"orientation":[],"userdata":""}}}