Flexible Physical Problem-Solving in Minds and Machines

Abstract: 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 efficiency and generalization. In this talk, I will discuss my work investigating the factorizations of objects, relations, and physics to support flexible physical problem-solving in both humans and machines. My research suggests that these ingredients can explain complex cognitive phenomena such as how people effortlessly learn to use new tools, and complex behaviours in machines such as highly realistic simulation and tool innovation. 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 while also better explaining how natural intelligence learns so much from so little.

Bio: Kelsey is currently a Senior Research Scientist at DeepMind. She received her PhD from MIT in the Computational Cognitive Science group, and her BSc from the University of British Columbia in physics. Her work has received awards including the international Glushko prize for best dissertation in cognitive science, a best paper award from Robotics: Science and Systems (R:SS), and an NSERC PhD fellowship. Spanning robotics, machine learning, and cognitive science, her work aims to elucidate the mechanisms that give rise to adaptive and efficient learning, especially in the domain of physical problem-solving.