THE INSTITUTE FOR NEURAL COMPUTATION

Annual Rockwood Memorial Lecture

What Should Computer Vision Learn from Neuroscience?

Tomaso Poggio

Center for Biological and Computational Learning

Computer Science and Artificial Intelligence Laboratory

McGovern Institute for Brain Research

Massachusetts Institute of Technology

4:00PM, Thursday, April 5, 2007

Leichtag Auditorium

Leichtag Family Biomedical Research Building, University of California, San Diego

The problem of learning is one of the main gateways to making intelligent

machines and to understanding how the brain works. In this talk I will give

a brief overview of recent efforts in my group developing machines that

learn for applications such as visual recognition and speech synthesis. I

will then review hierarchical feedforward quantitative models of the ventral

stream which, heavily constrained by physiology and biophysics, are

surprisingly successful in explaining several physiological data and

psychophysical results in scene categorization. I will then focus on the

limitations of such models for object recognition, pointing to questions

about the computational role of attention and about recognition tasks beyond

scene classification.

Host: Terry Sejnowski

The Rockwood Memorial Lectures are endowed by Mr. and Mrs. Jerome Rockwood

in memory of their late son, Paul, who received a B.S. in Computer Science

from UCSD in 1980 and then obtained a second degree B.A. in Psychology in

1981. In 1983 he stared a company, Integral Solutions, to develop a

universal language translation, but died tragically in a mountaineering

accident before he could fulfill his promise.