Friday, December 11, 2009
7.30 am - 10.30 am
16.00 am - 19.00 pm
Simon Haykin, McMaster University
Terry Sejnowski, Salk Institute and UCSD
Steven Zucker, Yale University
Tutorial Lectures (Haykin, Sejnowski)
Invited Speakers (Valiant, Tsotsos, Tishby, Kanerva, Tesauro)
The notion of "Curse of Dimensionality" was coined by Richard Bellman (1961). It refers to the exponential increase in computing a task of interest when extra dimensions are added to an associated mathematical space. For example, it arises in solving dynamic programming and optimal control problems when the dimension of the state vector is large. It also arises in solving learning problems when a finite number of data samples is used to learn a "state of nature, the distribution of which is infinitely large."
Much has been written on the curse of dimensionality problem in the mathematics and engineering literature. In contrast, little is known on how the human brain solves problems of this kind with relative ease. The key question is: How does the brain do it? To address this basic problem, it may be that we can learn from the mathematics and engineering literature, reformulated in the context of neuroscience.
This one-day workshop at NIPS 2009 is aimed at addressing the issues involved in the curses (and blessings) of dimensionality.
7:30AM Terry Sejnowski (Salk Institute)
Tutorial: Scaling Principles and Brain Architecture
7:45AM Simon Haykin (McMaster University)
Tutorial: The Curse of Dimensionality and How to Mitigate It in Dynamic Programming Applications
8:15AM Break
8:30AM Gerry Tesauro (IBM Watson)
RL Successes and Challenges in High-Dimensional Games
9:15AM Break
9:30AM John Tsotsos (York University)
How the Brain Deals with the Computational Complexity of Vision: A Different Kind of Dimensionality Curse
4:00PM Naftali Tishby (Weizmann Institute)
Predictive Information Bottleneck:
Why Simple Organisms Can Cope with Complex Environments"
4:45PM Break
5:00PM Les Valiant (Harvard University)
Experience-Induced Neural Circuits That Achieve High Capacity
5:45PM Break
6:00PM Pentti Kanerva(Stanford University)
Hyper-dimensional Computing: Computing in Distributed Representation with High-dimensional Random Vectors
Each speaker will have 45 minutes including discussion.
The talks are informal with interruptions welcome during the talks.