Speaker: David Cox
Location: Maxwell Dworkin G125, 33 Oxford Street, Cambridge, MA 02138
Time: Informal lunch with speaker, 12:30pm. Talk, 1:00pm
Title: Leveraging High-Performance Parallel Computing for Biologically Inspired Object and
Face Recognition
Abstract:
The visual cortex of the human brain is unrivaled by artificial systems in its ability to
recognize faces and objects in highly variable and cluttered real-world environments.
Biologically inspired computer vision systems seek to capture key aspects of the
computational architecture of the brain. Such approaches have proven successful across a
range of standard object- and face-recognition tasks. However, since the number of
parameters in a vision model is typically large, and the computational cost of evaluating
one particular parameter set is high, when a model fails we are left uncertain whether it
is because we are missing a fundamental idea, or because the correct "parts"
have not been tuned correctly, assembled at sufficient scale, or provided with enough
training. In this talk, I'll present a high-throughput search approach for exploring
a broad range of biologically inspired vision models. I'll discuss parallel
programming techniques that enable this approach, including machine-learning-guided
metaprogramming techniques that bring the ideas of high-throughput search down to the
level of implementation-level code optimization.
Bio:
David Cox is an assistant professor in the Department of Molecular and Cellular Biology
and the Center for Brain Science at Harvard. He completed his PhD in the Department of
Brain and Cognitive Sciences at MIT with a specialization in computational neuroscience.
Prior to joining MCB/CBS, he was a Junior Fellow at the Rowland Institute at Harvard, a
multidisciplinary institute focused on high-risk, high-reward scientific research at the
boundaries of traditional fields. His laboratory seeks to understand the computational
underpinnings of high-level visual processing through both reverse and forward
engineering. The group employs a wide range of experimental techniques to probe natural
systems, while at the same time actively developing practical computer vision systems
based on what is learned about the brain.
For information about future IACS events, see
http://iacs.seas.harvard.edu/events.
_______________________________________________
Iacs-events mailing list
Iacs-events(a)seas.harvard.edu
https://lists.seas.harvard.edu/mailman/listinfo/iacs-events