---------- Forwarded message ---------- From: Doina Precup dprecup@cs.mcgill.ca Date: Fri, Oct 30, 2009 at 9:26 AM Subject: Fwd: Reminder: Sridhar Mahadevan colloquium TODAY (Fir Oct. 30) 3:30pm, MC 13 To: Guillaume Desjardins guillaume.desjardins@gmail.com
Please forward to your seminar list
Thanks, Doina
Begin forwarded message:
From: Doina Precup dprecup@cs.mcgill.ca Date: October 30, 2009 9:19:51 AM EDT (CA) To: profs@cs.mcgill.ca, grads@cs.mcgill.ca, labrl@cs.mcgill.ca Subject: Reminder: Sridhar Mahadevan colloquium TODAY (Fir Oct. 30) 3:30pm, MC 13
2009/10/30, MC13, 15:30 - 16:30
New Frontiers in Representation Discovery Sridhar Mahadevan, University of Massachusetts-Amherst
Abstract: A longstanding challenge in artificial intelligence is how to automatically construct representations by analyzing state spaces. This problem has attracted renewed interest in a variety of areas, including compression and dimensionality reduction, planning and reinforcement learning, and semi-supervised and unsupervised learning. This talk will describe a family of spectral approaches to constructing new representations by diagonalizing or dilating Laplacian operators over a state space. A number of challenges and tradeoffs will be discussed, including task-specific vs. task-invariant representations; flat vs. multi-scale approaches; orthogonal vs. redundant bases; and dealing with the curse of dimensionality. A range of case studies from our recent research will be presented, including a paradigm for solving Markov decision processes where representation and control are learned simultaneously; a spectral method for clustering of text documents where the topic hierarchy is automatically constructed; and a compression method for computer graphics based on multiscale analysis of object geometry.
Biography of Speaker: Sridhar Mahadevan is Co-Director of the Autonomous Learning Laboratory at the Department of Computer Science, University of Massachusetts, Amherst. His research interests are in artificial intelligence and machine learning. He is an associate editor of the Journal of Machine Learning Research, and was a tutorial speaker at AAAI 2007, IJCAI 2007, and ICML 2006. In 2008, he authored a book on representation discovery as part of the Synthesis Lectures in AI and Machine Learning series published by Morgan Claypool.
As usual, refreshments will be served in the SOCS lounge after the talk.
Hope to see you there! Feel free to forward to others who may be interested Doina