Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session C: Friday, August 15, 2:00 – 5:00 pm, de Brug & E‑Hall

A semantic bias for accurate predictive learning in multidimensional environments

Euan Prentis¹, Akram Bakkour; ¹University of Chicago

Presenter: Euan Prentis

To make accurate inferences about our multidimensional world, humans must distinguish observations of causal processes from spurious associations. We investigated the role of inductive biases in shaping memory around causal information, specifically testing for a semantic bias that leverages existing semantic structure to direct learning. Participants completed a predictive learning task in which both causal and spurious associations were observed. Results showed that spurious inferences were suppressed when the causal associations were defined within semantic categories, indicating that a semantic bias directed learning. Simulations of a feature-based successor features model further demonstrated that this bias should have a more dramatic benefit in more naturalistic environments, with high-dimensional states and deep causal processes. In all, this work demonstrates that inductive biases that act on multidimensional transition dynamics may be essential for learning in our complex world.

Topic Area: Memory, Spatial Cognition & Skill Learning

Extended Abstract: Full Text PDF