BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Drupal iCal API//EN X-WR-CALNAME:Events items teaser X-WR-TIMEZONE:America/Toronto BEGIN:VTIMEZONE TZID:America/Toronto X-LIC-LOCATION:America/Toronto BEGIN:DAYLIGHT TZNAME:EDT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 DTSTART:20200308T070000 END:DAYLIGHT BEGIN:STANDARD TZNAME:EST TZOFFSETFROM:-0400 TZOFFSETTO:-0500 DTSTART:20191103T060000 END:STANDARD END:VTIMEZONE BEGIN:VEVENT UID:69f40b55b1b4a DTSTART;TZID=America/Toronto:20200917T160000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200917T160000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/phd-seminar-p roblems-and-opportunities-training-deep LOCATION:Online 200 University Avenue West Waterloo ON N2L 3G1 Canada SUMMARY:PhD Seminar: Problems and Opportunities in Training Deep Learning\n Software Systems: An Analysis of Variance CLASS:PUBLIC DESCRIPTION:PLEASE NOTE: THIS PHD SEMINAR WILL BE GIVEN ONLINE.\n\nHUNG PHA M\, PHD CANDIDATE\n_David R. Cheriton School of Computer Science_\n\nSUPE RVISORS: Professors Lin Tan and Yaoliang Yu DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b2d1a DTSTART;TZID=America/Toronto:20200319T103000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200319T113000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/ai-seminar-ze ro-shot-learning-generalized-information LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 1304 Waterloo ON N2L 3G1 Canada SUMMARY:AI Seminar: Zero-Shot Learning: Generalized Information Transfer\nA cross Classes CLASS:PUBLIC DESCRIPTION:YUHONG GUO\, SCHOOL OF COMPUTER SCIENCE\n_Carleton University_\ n\nThe need for annotated data is a fundamental bottleneck in developing\n automated prediction systems. A key strategy for reducing the reliance\non human annotation is to exploit generalized information transfer\,\nwhere a limited data resource is augmented with labeled data collected\nfrom rel ated sources.  DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b3951 DTSTART;TZID=America/Toronto:20200402T103000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200402T113000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/cancelled-ai- seminar-graph-guided-predictions LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 1304 Waterloo ON N2L 3G1 Canada SUMMARY:CANCELLED • AI Seminar: Graph Guided Predictions CLASS:PUBLIC DESCRIPTION:VIKAS GARG\, ELECTRICAL ENGINEERING & COMPUTER SCIENCE\n_Massac husetts Institute of Technology_\n\nIn this talk I will describe our recen t work on effectively using\ngraph structured data. Specifically\, I will discuss how to compress\ngraphs to facilitate predictions\, understand the capacity of\nalgorithms operating on graphs\, and how to infer interactio n graphs so\nas to predict deliberative outcomes. DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b44fa DTSTART;TZID=America/Toronto:20200312T103000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200312T113000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/ai-seminar-de ep-learning-graphs LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 1304 Waterloo ON N2L 3G1 Canada SUMMARY:AI Seminar: Deep Learning on Graphs CLASS:PUBLIC DESCRIPTION:RENJIE LIAO\, DEPARTMENT OF COMPUTER SCIENCE\n_University of To ronto_\n\nGraphs are ubiquitous in many domains like computer vision\, nat ural\nlanguage processing\, computational chemistry\, and computational so cial\nscience. Although deep learning has achieved tremendous success\,\ne ffectively handling graphs is still challenging due to their discrete\nand combinatorial structures. In this talk\, I will discuss my recent\nwork w hich improves deep learning on graphs from both modeling and\nalgorithmic perspectives. DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b501d DTSTART;TZID=America/Toronto:20200309T103000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200309T113000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/ai-seminar-ov ercoming-mode-collapse-and-curse-dimensionality LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 1304 Waterloo ON N2L 3G1 Canada SUMMARY:AI Seminar: Overcoming Mode Collapse and the Curse of Dimensionalit y CLASS:PUBLIC DESCRIPTION:KE LI\, RESEARCH SCIENTIST\n_Google_\n\nIn this talk\, I will p resent our work on overcoming two long-standing\nproblems in machine learn ing and computer vision DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b596f DTSTART;TZID=America/Toronto:20200203T103000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200203T103000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/ai-seminar-co sts-and-benefits-invariant-representation LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 1302 Waterloo ON N2L 3G1 Canada SUMMARY:AI Seminar: Costs and Benefits of Invariant Representation Learning CLASS:PUBLIC DESCRIPTION:HAN ZHAO\, MACHINE LEARNING DEPARTMENT\n_Carnegie Mellon Univer sity_\n\nThe success of supervised machine learning in recent years crucia lly\nhinges on the availability of large-scale and unbiased data\, which i s\noften time-consuming and expensive to collect. Recent advances in deep\ nlearning focus on learning invariant representations that have found\nabu ndant applications in both domain adaptation and algorithmic\nfairness. Ho wever\, it is not clear what price we have to pay in terms\nof task utilit y for such universal representations. In this talk\, I\nwill discuss my re cent work on understanding and learning invariant\nrepresentations.  DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b6475 DTSTART;TZID=America/Toronto:20200127T103000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200127T103000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/ai-seminar-co ld-start-universal-information-extraction LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 1304 Waterloo ON N2L 3G1 Canada SUMMARY:AI Seminar: Cold-Start Universal Information Extraction CLASS:PUBLIC DESCRIPTION:LIFU HUANG\, DEPARTMENT OF COMPUTER SCIENCE\n_University of Ill inois at Urbana–Champaign_\n\nWho? What? When? Where? Why? are fundament al questions asked when\ngathering knowledge about and understanding a con cept\, topic\, or\nevent. The answers to these questions underpin the key information\nconveyed in the overwhelming majority\, if not all\, of langu age-based\ncommunication. Unfortunately\, typical machine learning models and\nInformation Extraction (IE) techniques heavily rely on human annotate d\ndata\, which is usually very expensive and only available and compiled\ nfor very limited types or languages\, rendering them incapable of\ndealin g with information across various domains\, languages\, or other\nsettings . DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b6d7c DTSTART;TZID=America/Toronto:20200124T130000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200124T130000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/ai-seminar-al oha-artificial-learning-human-attributes LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 2306C Waterloo ON N2L 3G1 Canada SUMMARY:AI Seminar: ALOHA: Artificial Learning of Human Attributes for\nDia logue Agents CLASS:PUBLIC DESCRIPTION:STEVEN Y. FENG\n_David R. Cheriton School of Computer Science_\ n\nFor conversational AI and virtual assistants to communicate with\nhuman s in a realistic way\, they must exhibit human characteristics\nsuch as ex pression of emotion and personality. Current attempts toward\nconstructing human-like dialogue agents have presented significant\ndifficulties.  DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b78ab DTSTART;TZID=America/Toronto:20200120T103000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20200120T103000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/ai-seminar-ri sk-aware-machine-learning-scale LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 1304 Waterloo ON N2L 3G1 Canada SUMMARY:AI Seminar: Risk-Aware Machine Learning at Scale CLASS:PUBLIC DESCRIPTION:JACOB GARDNER\, RESEARCH SCIENTIST\n_Uber AI Labs_\n\nIn recent years\, machine learning has seen rapid advances with\nincreasingly large scale and complex data modalities\, including\nprocessing images\, natura l language and more. As a result\,\napplications of machine learning have pervaded our lives to make them\neasier and more convenient. Buoyed by thi s success\, we are approaching\nan era where machine learning will be used to autonomously make\nincreasingly risky decisions that impact the physic al world and risk\nlife\, limb\, and property. DTSTAMP:20260501T020925Z END:VEVENT BEGIN:VEVENT UID:69f40b55b83c6 DTSTART;TZID=America/Toronto:20191206T110000 SEQUENCE:0 TRANSP:TRANSPARENT DTEND;TZID=America/Toronto:20191206T110000 URL:https://uwaterloo.ca/artificial-intelligence-group/events/phd-seminar-a ddressing-labels-shortage-segmentation-3 LOCATION:DC - William G. Davis Computer Research Centre 200 University Aven ue West 3102 Waterloo ON N2L 3G1 Canada SUMMARY:PhD Seminar: Addressing Labels Shortage: Segmentation with 3%\nSupe rvision CLASS:PUBLIC DESCRIPTION:DMITRII MARIN\, PHD CANDIDATE\nDavid R. Cheriton School of Comp uter Science\n\nDeep learning models generalize limitedly to new datasets and require\nnotoriously large amounts of labeled data for training. The l atter\nproblem is exacerbated by the need of ensuring that trained models are\naccurate in large variety of image scenes. The diversity of images\nc omes from combinatorial nature of real world scenes\, occlusions\,\nvariat ions in lightning\, acquisition methods\, etc. Many rare images\nmay have little chance to be included in a dataset\, but are still very\nimportant\ , as they often represent situations where a recognition\nmistake has a hi gh cost. DTSTAMP:20260501T020925Z END:VEVENT END:VCALENDAR