Structure Learning in Deep Learning | HKUST CSE

                Joint Seminar
The Hong Kong University of Science & Technology
Department of Computer Science and Engineering
Big Data Institute
Human Language Technology Center
Speaker:        Prof. Pascal Poupart
                Professor, David R. Cheriton School of Computer Science
                University of Waterloo

Title:          "Structure Learning in Deep Learning"

Date:           Monday, 20 November 2017

Time:           4:00pm - 5:00pm

Venue:          Lecture Theater F (near lift 25/26),HKUST


The goal of data science is to extract insights from unstructured and
complex data. This often hinges on the use of a good representation where
suitable features can simplify tremendously the extraction of knowledge.
Traditionally, features were handcrafted based on domain knowledge.
However, recent advances in deep learning have shown that it is often
better to use a deep structure in which features are automatically learned
from data. This has completely revolutionized computer vision, speech
recognition and natural language. That being said, feature engineering is
now replaced by architecture engineering since practitioners spend
enormous time adjusting the architecture and hyperparameters by trial and
error. Hence there is a need for techniques to automatically learn the
structure and hyperparameters of networks. In this talk, I will show how
to automatically learn the structure of a special class of deep neural
networks known as sum-product networks from streaming data. This will be
demonstrated in variety of domains where it is unclear what architecture
might work well.


Pascal Poupart 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
Assistive Technologies, Natural Language Processing and Telecommunication
Networks. 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 real-world problems,
including automated prompting for people with dementia for the task of
handwashing and spoken dialog management. Other notable projects that his
research team are currently working on include deep learning with clear
semantics, structure learning, personalized transfer learning,
conversational agents, adaptive satisfiability and stress detection based
on wearable devices.

Pascal Poupart received a Cheriton Faculty Fellowship (2015-2018), a best
student paper honourable mention (SAT-2017), an outstanding collaborator
award from Huawei Noah's Ark (2016), a top reviewer award (ICML-2016), the
best main track solver and best application solver (SAT-2016 competition),
a best reviewer award (NIPS-2015), an Early Researcher Award from the
Ontario Ministry of Research and Innovation (2008), two Google research
awards (2007-2008), a best paper award runner up (UAI-2008) and the IAPR
best paper award (ICVS-2007). He also serves as associate editor of the
Journal of Artificial Intelligence Research (JAIR) (2017 - present),
member of the editorial board of the Journal of Machine Learning Research
(JMLR) (2009 - present) and guest editor for Machine Learning Journal
(MLJ) (2012 - present). He routinely serves as area chair or senior
program committee member for NIPS, ICML, AISTATS, IJCAI, AAAI and UAI. He
serves as technical advisor for Huawei Technologies, ElementAI, TalkIQ and
ProNavigator. His research collaborators include Huawei Technologies,
Google, Intel, Kik Interactive, In the Chat, Slyce, HockeyTech,
ProNavigator, the Alzheimer Association, the UW-Schlegel Research
Institute for Aging, Sunnybrook Health Science Centre and the Toronto
Rehabilitation Institute.