Jessica Thompson

To respond adaptively in novel settings, intelligent agents must develop certain invariances (or abstractions) to map the infinite variety of the natural world to some smaller number of concepts, features, or values. Dr. Jessica Thompson investigates the computational principles that govern how those abstract representations develop through experience.

As a postdoctoral researcher in Chris Summerfield’s group at University of Oxford, she uses careful experimental design to establish the functional role of various computational ingredients in artificial neural networks trained on visual reasoning tasks. She completed her PhD at the International Laboratory for Brain, Music & Sound Research (BRAMS) and the Quebec Artificial Intelligence Institute (Mila) at University of Montreal where her research focused on characterizing and comparing representational hierarchies in artificial neural networks and the human auditory system.