Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster E in Poster Session E: Thursday, August 6, 10:30 am – 12:15 pm, Kimmel Center, Shorin & Rosenthal Rooms

Belief Filtering: Meta-inference of Internal Models

Aniruddh R Galgali¹, Maneesh Sahani²; ¹University College London, ²Gatsby Unit, UCL

Presenter: Aniruddh R Galgali

Characterising the internal states underlying behaviour is challenging. Human participants or other "agents" may act rationally, but with beliefs shaped by mistaken models of the causal structure of the environment and by internal noise. External "observers" may access a veridical world state and witness behaviour, but have no direct view of beliefs, noise, or the internal model that shapes them. Here, we demonstrate a formal framework to clearly distinguish agent and observer models, and their belief states, within an amortised meta-inference approach. The approach makes it possible to track the evolution of an agent's beliefs across time, and thus identify the internal model on which it acts.

Topic Area: Methods, Tools, Theory & Neural Coding