Arts & Sciences Events
[PAST EVENT] Colloquium talk: Graph Structure as A Double-Edged Sword in Machine Learning
Location
McGlothlin-Street Hall, Online on Zoom251 Jamestown Rd
Williamsburg, VA 23185Map this location
Abstract:
Graph structures describing relationships and interactions among
entities are everywhere, ranging from chemical molecule structures in
??microworld, societal networks in daily life to the Internet to
defender-adversary interactions in national security. The ubiquity of
such a graph-structured description of our world calls for effective and
trustworthy methods that make use of and learn to understand information
represented in this structured form. On one hand, graph structure brings
relational inductive bias to facilitate learning algorithms; on the
other hand, such structure could be used in an undesired way threatening
the trustworthiness of machine learning.
In this talk, I will introduce my work on computational questions
pertaining to the double-edged role of graph structure in machine
learning. In particular, I will mainly illustrate the opportunities and
the trustworthiness issues through three threads of my research: 1) how
to utilize graph structure to enhance machine learning when label
information is limited; 2) how to mitigate societal discrimination in
predictions resulting from structural bias; 3) how to understand the
robustness of structure when it can be altered by an adversary. I will
demonstrate my studies on these questions lying at the intersection of
machine learning and graph theory.
Bio:
Lu Lin is a Ph.D. candidate in the Computer Science Department at the
University of Virginia (UVa), advised by Professor Hongning Wang. Before
coming to UVa, she obtained her B.S. and M.S. degree in Computer Science
Department from Beihang University. Her research interests include
artificial intelligence, data mining and applied machine learning on
graph-structured data. She focuses on utilizing graph structure to
advance machine learning in a trustworthy and responsible manner. Her
research has been published at high-impact venues (e.g., KDD, WWW,
AISTATS, WSDM, TKDE).
Contact
Pieter Peers