Towards a recipe for physical reasoning in humans and machines

The world is structured in countless ways. When cognitive and machine models respect these structures, by factorizing their modules and parameters, they can achieve remarkable accuracy and generalization. In this talk, I will discuss our work investigating the factorizations of objects, relations, and physics in both humans and machines. Focusing on problems in physical reasoning including construction and tool use, I will show how to harness object and relational structure in the form of graph networks and probabilistic programs to improve machine generalization and explain how people can rapidly acquire new tool-using strategies, like the notion of a catapult, from just a handful of experiences. By taking better advantage of problem structure, and combining it with general-purpose methods for statistical learning, we can develop more robust and data-efficient machine agents, and better explain how humans learn so much from so little.

Zoom details: https://us04web.zoom.us/j/73872751035?pwd=M1dMZ3V0TFQ5NDVrQU9TYnY3OU9KUT09

ID de réunion : 738 7275 1035
Code secret : L0MFXm

This seminar aims to open that dialogue between the fields of Cognitive Science and Artificial Intelligence. Speakers may come from one field or the other, but all will use this opportunity to reflect on how a pairing between the two fields can be stronger than the sum of the parts.

https://cogai-seminar.github.io/