[Research] Bert Huang's talk is today at noon in NSH 3001

[Jeff Schneider]

Please come and listen at noon today in NSH 3001!!

Talk Title: Learning with Degree-based Subgraph Estimation

Abstract: Networks and their topologies are critical to nearly every aspect of 
modern life, with social networks governing human interactions and computer 
networks governing global information-flow. Network behavior is inherently 
structural, and thus modeling data from networks should benefit from explicitly 
modeling structure. This talk will cover methods for and analysis of learning 
from network data while explicitly modeling one important measure of structure: 
degree. Central to this work is a procedure for exact maximum likelihood 
estimation of a distribution over graph structure, where the distribution 
factorizes into edge-likelihoods for each pair of nodes and degree-likelihoods 
for each node. The estimation can be solved using distributable belief 
propagation, for which we provide various scalability improvements. We 
additionally propose a learning algorithm for learning the parameters of the 
distribution from network data consisting of node attributes and network 
connectivity, using strategies similar to max-margin structured prediction. We 
apply the learning algorithm and predictor to real network data from online 
social and document networks and are able to better predict networks than 
various baselines.

Speaker Bio: Bert Huang is a Ph.D. candidate studying machine learning at 
Columbia University. He is a member of the Machine Learning Laboratory directed 
by Tony Jebara and the Center for Computational Learning Systems (CCLS). He 
received his MS in 2006 from Columbia University and his BS and BA in 2004 from 
Brandeis University. Bert's papers on topics including belief propagation, 
collaborative filtering, and graph prediction have appeared at conferences 
including The International Conference on Artificial Intelligence and Statistics 
(AISTATS) and the International Conference on Machine Learning and Applications 
(ICMLA). Between 2006 and 2010, Bert was part of a collaboration between CCLS 
and Consolidated Edison, the primary energy provider of New York City, using 
machine learning to guide decision-making in the field, leading to several 
publications, including a paper to appear in IEEE Transactions on Pattern 
Analysis and Machine Intelligence (TPAMI). Bert was a research intern at IBM 
Research Watson in the summer of 2010, working on analysis of spatiotemporal 
city data. Between 2008 and 2010, Bert served as a preceptor (student lecturer) 
within Columbia's Department of Computer Science, for which the Department 
awarded him the Andrew P. Kosoresow Memorial Award for Outstanding Performance 
in TA-ing and Service. Bert's research interests within and surrounding machine 
learning include network analysis, probabilistic modeling of networks, message 
passing inference, social media, and, most recently, large-scale machine learning.