HKBU  |  SCI  |  BUniPort  |  Library  |  Alumni  |  Job Vacancies  |  Intranet  |  Sitemap        
Undergraduate Admissions
Taught Postgraduate Admissions
Research Postgraduate Admissions
Scholarship
News & Achievements
Events
Video
Research Highlights
Contact & Direction
International Exchange and Internship Programmes
 
HONG KONG BAPTIST UNIVERSITY
FACULTY OF SCIENCE

Department of Computer Science Colloquium
2019 Series

On Optimizing Inference in Probabilistic Graphical Models

Prof. Cory J. Butz
Associate Dean (Research)
Faculty of Science
University of Regina
Canada

Date: August 7, 2019 (Wednesday)
Time: 11:00 am - 12:00 pm
Venue: SCT909, Cha Chi Ming Science Tower, Ho Sin Hang Campus

Abstract
Probabilistic graphical models (PGMs), a marriage of probability theory and graph theory, have long played an important role in managing uncertainty in artificial intelligence. Bayesian networks, in particular, are perhaps the most well-known PGM. Join tree propagation (JTP) is one way to conduct Bayesian network inference and central to the theory and practice of probabilistic expert systems. We present the state-of-the-art JTP algorithm, called Simple Propagation. As the name suggests, Simple Propagation performs Bayesian network inference without the overhead found in other competing methods. Experimental results on benchmark datasets show that Simple Propagation tends to be the fastest JTP algorithm. Simple Propagation arose out of our earlier work on developing a clever way to graphically represent probability information. Looking forward, JTP can be used to construct a new type of PGM known as Sum-Product Networks (SPNs). SPNs are a deep learning model with tractable inference, a favorable property compared to Bayesian networks. We show how exact inference can be performed in SPNs without necessarily visiting every node in the SPN. This observation was based upon an optimization technique applied in Bayesian network inference. Lastly, we present Deep Convolutional Sum-Product Networks, which formally establish when a convolutional neural network defines an SPN.

Biography
Cory J. Butz received his Ph.D. degree in computer science from the University of Regina, Regina, SK, Canada, in 2000. He then joined the School of Information Technology and Engineering at the University of Ottawa, Ottawa, ON, Canada, as an Assistant Professor. In 2001, he returned to the Department of Computer Science at the University of Regina, where he now holds the rank of Professor and serves as the Associate Dean (Research) in the Faculty of Science. He has served as President of the Canadian Artificial Intelligence Association (CAIAC). His research findings on probabilistic graphical models have drawn invitations to visit Google, MIT, and the University of Cambridge.

********* ALL INTERESTED ARE WELCOME ***********
(For enquiry, please contact Computer Science Department at 3411 2385)

http://www.comp.hkbu.edu.hk/v1/?page=seminars&id=533
Photos  Slides
Copyright © 2019. All rights reserved.Privacy Policy
Department of Computer Science, Hong Kong Baptist University
Hong Kong Baptist University