Kris Jensen did his PhD with Guillaume Hennequin at Cambridge and is now a postdoc in the lab of Tim Behrens.
You can read more about Kris’ fascinating work here.

Do come along if you’re interested in how brain builds and uses models of the world!

Title
A spacetime attractor model of planning in frontal cortex

Abstract

Animals can flexibly navigate complex and changing environments, and this flexibility is facilitated by an internal world model that supports planning. Prefrontal cortex (PFC) has commonly been implicated in such model-based planning, but little is known about the underlying algorithms and neural mechanisms. Inspired by recent findings of explicit future representations for working memory, we develop a new mechanistic theory of model-based planning in PFC. We propose that PFC contains a neural attractor network, where the fixed points of the network dynamics are ‘paths’ through space and time that maximize expected reward. This allows for planning to proceed as an inference process over future states, conditioned on a current state and reward function. The result of planning is a representation where separate neural subspaces represent the expected state of the world at different times along the planned trajectory. We show that this algorithm is naturally learned by RNNs trained on challenging planning tasks, and it generalizes to domains where previous mechanistic models fail.

Time and place
CFIN South, Thalamus, Wednesday 28/5, 09:30-10:30.

ALL ARE WELCOME.