From Lovasz theta to kernel methods: A new connection between graph theory and machine learning

Lecturer : 
Vinay Jethava, Chalmers University of Technology
Event type: 
HIIT seminar
Event time: 
2014-02-10 13:15 to 14:00
Aalto University, Computer Science Building, lecture hall T2


A number of machine learning problems involve analysis of graphs and have borrowed extensively from graph theory - spectral clustering, graph kernels, etc. In this talk, I’ll present our recent work on establishing a connection between Lovasz theta function, a powerful concept in graph theory which has been used heavily in algorithms and combinatorial optimization;  and kernel methods in machine learning. This connection has several implications including design of a fast algorithm for common dense subgraph detection — a problem arising in integrative analysis of  microarray datasets.

About the speaker:

Vinay is a fourth year doctoral student in Computer Science at Chalmers University, Gothenburg, working under the supervision of Devdatt Dubhashi.  His research is focussed on integrative analysis of multiple networks using machine learning techniques such as probabilistic graphical models and kernel methods. Before coming to Chalmers, he obtained his MS (CS) at Indian Institute of Science (IISc) and Bachelors (EE) at Indian Institute of Technology-Madras (IITM) with a stint as a VLSI engineer in between.

More about his work can be found at:

Last updated on 4 Feb 2014 by Antti Ukkonen - Page created on 4 Feb 2014 by Antti Ukkonen