EE Seminar: VBNets: Learning Entity Representations via Variational Bayesian Networks
(The talk will be given in English)
Speaker: Dr. Oren Barkan
Microsoft, Israel
SUNDAY, January 12th, 2020
15:00 - 16:00
Room 011, Kitot Bldg., Faculty of Engineering
VBNets: Learning Entity Representations via Variational Bayesian Networks
Abstract
Learning entity representations is an active research field. In the last decade, both the NLP and recommender systems communities introduced a plethora of methods for mapping words, items and users to vectors in a latent space. The vast majority of these works utilize implicit co-occurrences relations (e.g. co-occurrences of words in text, co-consumption of items by users) for learning the latent entity vectors. Yet, often, additional side information in the form of explicit (e.g. hierarchical, semantic, syntactic) relations can be leveraged for learning finer embeddings.
In this talk, we present Variational Bayesian Networks (VBNets) - A novel scalable hierarchical Bayesian model that utilizes both implicit and explicit relations for learning entity representations. VBNets are designed for Microsoft Store and Xbox services that handle around a billion users worldwide. Different from point estimate solutions that map entities to vectors and are usually over confident, VBNets map entities to densities in the latent space and hence model uncertainty. VBNets are based on analytical approximations of the intractable entities' posterior and the posterior predictive distribution of the data. We demonstrate the effectiveness of VBNets on linguistic, recommendations, and medical informatics tasks, where it is shown to outperform other alternative methods that facilitate Bayesian modeling with or without semantic priors. In addition, we show that VBNets produce superior representations for rare words and cold items. If time permits, we will give a brief overview of several recent deep learning works in the domains of deep neural attention mechanisms, multiview representation learning and inverse problems with applications for natural language understanding, recommender systems, computer vision, sound synthesis and biometrics.
Short Bio
Oren Barkan is a Principal Researcher at Microsoft, where he was previously a post-doctoral researcher, collaborating with Microsoft Research UK and Microsoft Israel. Prior to that, he was with Google Research and IBM Research. He received his Ph.D. from Tel Aviv University, under the supervision of Prof. Amir Averbuch. His research interests are deep neural attention mechanisms, representation learning, multiview learning, Bayesian inference and inverse problems with applications for computer vision, natural language understanding, recommender systems, speech analysis, sound synthesis, biometrics, inflation forecasting, healthcare and medical informatics. He is the author of more than 40 research papers and patents.
