Résumé: Most problems in Artificial Intelligence are NP-hard and therefore a significant amount of work has been devoted to exploit problem-specific structure or design approximate algorithms with good scalability properties. With the rise of data science, a paradigm shift is taking place. Instead of having a domain expert specify a model for which reasoning is intractable, machine learning allows us to learn tractable models directly from data. The reality is that most models specified by domain experts are approximate while models obtained from data can be more accurate. Furthermore, by employing a hierarchy of models that equate representation complexity with reasoning complexity, it is possible to learn models whose complexity increases with the amount of data and therefore remain tractable. For example, inference in probabilistic graphical models such as Bayesian networks and Markov networks is #P-complete. However if we estimate sum-product networks (special type of deep neural networks that are equivalent to Bayesian and Markov networks) directly from data, exact probabilistic inference can be done in linear time with respect to the size of the network. In this talk, I will explain how to learn sum-product networks from large streaming datasets in an online and distributed fashion for a wide range of applications including language modeling, handwriting recognition, recommender systems, communication networks, classification and time-series prediction. I will also explain how to learn extensions of sum-product networks for tractable decision making in various diagnosis tasks.
This presentation will be based on the following articles:
Han Zhao, Mazen Meliberi, Pascal Poupart, On the Relationship Between Sum-Product Networks and Bayesian Networks, ICML, 2015.
Abdullah Rashwan, Han Zhao and Pascal Poupart, Online and Distributed Bayesian Moment Matching for Parameter Learning in Sum-Product Networks, AISTATS, 2016.
Mazen Melibari, Pascal Poupart and Prashant Doshi, Sum-Product-Max Networks for Tractable Decision Making, IJCAI, 2016
Mazen Melibari, Pascal Poupart, Prashant Doshi, George Trimponias, Dynamic Sum-Product Networks for Tractable Inference on Sequence Data, PGM, 2016.
Priyank Jaini, Abdullah Rashwan, Han Zhao, Yue Liu, Ershad Banijamali, Zhitang Chen and Pascal Poupart, Online Algorithms for Sum-Product Networks with Continuous Variables, PGM, 2016.
Han Zhao and Pascal Poupart, A Unified Approach for Learning the Parameters of Sum-Product Networks, NIPS, 2016.
Biographie: Pascal Poupart is an Associate Professor in the David R. Cheriton School of Computer Science at the University of Waterloo, Waterloo (Canada). He received the B.Sc. in Mathematics and Computer Science at McGill University, Montreal (Canada) in 1998, the M.Sc. in Computer Science at the University of British Columbia, Vancouver (Canada) in 2000 and the Ph.D. in Computer Science at the University of Toronto, Toronto (Canada) in 2005. His research focuses on the development of algorithms for reasoning under uncertainty and machine learning with application to health informatics and natural language processing. He is most well known for his contributions to the development of approximate scalable algorithms for partially observable Markov decision processes (POMDPs) and their applications in assistive technologies, including automated prompting for people with dementia for the task of handwashing. Other notable projects that his research team are currently working on include chatbots for automated personalized conversations and wearable analytics to assess modifiable health risk factors. He co-founded Veedata, a startup that provides analytics services to the insurance industry and the research market. Pascal Poupart received a David R. Cheriton Faculty Award in 2015 and an Early Researcher Award (competitive honor for top Ontario researchers) by the Ontario Ministry of Research and Innovation in 2008. He was also a co-recipient of the Best Paper Award Runner Up at the 2008 Conference on Uncertainty in Artificial Intelligence (UAI) and the IAPR Best Paper Award at the 2007 International Conference on Computer Vision Systems (ICVS). He also serves on the editorial board of the Journal of Machine Learning Research (JMLR) (2009 – present) and the Journal of Artificial Intelligence Research (JAIR) (2008 – 2011). His research collaborators include Huawei, Google, Intel, Kik Interactive, In the Chat, Slyce, HockeyTech, the Alzheimer Association, the UW-Schlegel Research Institute for Aging, Sunnybrook Health Science Centre and the Toronto Rehabilitation Institute.
Note: La présentation sera donnée en français.