Towards artificial general intelligence: Brain-inspired CAPTCHA breaking and Atari playing

Seminar | January 24 | 12-1 p.m. | 560 Evans Hall

 Miguel Lázaro-Gredilla, Vicarious, Inc.

 Neuroscience Institute, Helen Wills

Abstract: Compositionality, generalization, and learning from a few examples are among the hallmarks of human intelligence. In this talk I will describe how Vicarious combines these ideas to create approaches to CAPTCHA breaking and Atari game playing that improve on the state of the art. Both of these tasks have indeed been tackled before, using respectively Convolutional Neural Networks (CNNs) and the Asynchronous Advantage Actor-Critic (A3C). Despite the good results obtained using those techniques, their behavior is somewhat underwhelming when compared with what we can expect from humans in terms of generalization ability, resistance to noise and low sample complexity. I will explain how our models (the Recursive Cortical Network and Schema Networks) can improve on those by drawing inspiration from the human brain.

For more info: https://www.vicarious.com/blog/

 nrterranova@berkeley.edu