log in  |  register  |  feedback?  |  help  |  web accessibility
Logo
PhD Proposal: A Neurocomputational Model of Causal Reasoning and Compositional Working Memory for Imitation Learning
Gregory Davis
Virtual
Friday, April 17, 2020, 12:00-2:00 pm Calendar
  • You are subscribed to this talk through .
  • You are watching this talk through .
  • You are subscribed to this talk. (unsubscribe, watch)
  • You are watching this talk. (unwatch, subscribe)
  • You are not subscribed to this talk. (watch, subscribe)
Abstract
Causal reasoning and compositionality are increasingly recognized as important aspects of human cognition that are difficult to capture in neurocomputational models. Modest success has been achieved with the use of hybrid neuro-symbolic systems that combine artificial neural networks with auxiliary symbolic models, demonstrating that neural networks struggle with and benefit from robust control structures afforded by symbolic computation. This discrepancy is puzzling given the cognitive capabilities of human nervous systems, and indicates a computational explanatory gap between cognitive and neurocomputational algorithms that stifles efforts to implement human-level cognition in purely neural systems. Recent work has shown that programmable attractor-based neural networks are capable of executing cognitive procedures such as top-down control of working memory, attentional direction, and motor engagement, suggesting that these networks are theoretically capable of higher-level cognitive abilities such as causal reasoning with compositional knowledge. To this end, I propose to develop biologically and cognitively plausible methods for compositional working memory and causal reasoning for robotic imitation learning using programmable attractor networks. Given the cognitive and developmental significance of causal reasoning, compositional working memory, and imitation learning in human intelligence, this will contribute not only to the development of human-level neurocognitive systems, but also to elucidating the neurocomputational basis of cognition and bridging the computational explanatory gap. Preliminary work demonstrates that attractor networks can reliably store and manipulate graph-based data structures using compositional programs, and that they are viable for implementing efficient algorithms for cause-effect reasoning during imitation learning.

Examining Committee: 
 
                          Chair:               Dr. James Reggia
                          Dept rep:         Dr.  Mihai Pop
                          Members:        Dr. Yiannis Aloimonos
                                                
                                                    

 
Bio

Gregory Davis is a fourth year computer science PhD student at the University of Maryland. His research interests lie at the intersection of artificial intelligence and the neural and cognitive sciences, and he aims to develop biologically-inspired neurocomputational models that capture key aspects of human-level cognition such as compositional representation and hypothetico-deductive causal reasoning.

This talk is organized by Tom Hurst