Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Graduate Student Award Winner 

Cerebellar Cognition: Cerebellar Prediction Errors in Reinforcement and Statistical Learning

Poster Session A - Saturday, March 7, 2026, 3:00 – 5:00 pm PST, Fairview/Kitsilano Ballroom
Also presenting in Data Blitz Session 4 - Saturday, March 7, 2026, 10:30 am – 12:00 pm PST, Salon F.

Juliana E. Trach¹ (), Yiran Ou¹, Samuel D. McDougle^1,2; ¹Department of Psychology, Yale University, ²Wu Tsai Institute, Yale University

A prominent theory of cerebellar learning posits that the cerebellum acts as an internal model for our motor system, generating predictions about action outcomes and using prediction errors to refine actions. Cerebellar circuitry associated with prediction-error-based learning has been extensively described in the motor domain. Yet, whether this function extends to non-motor learning is unclear. Here, we investigate cerebellar encoding of prediction errors in two forms of non-motor learning: reinforcement learning (RL) and statistical learning (SL). Participants executed an RL (N = 32) and an SL (N = 20) task while undergoing fMRI. During RL, participants won points by choosing between stimuli with different reward probabilities (a standard “two-arm bandit” task). During SL, participants passively viewed a sequence of individually presented fractals, where certain fractal-fractal transitions were more likely than others. We manipulated the delay (.8s or 3s) between participant choices and reward feedback during RL and between fractal presentations during SL to test the hypothesis that cerebellar involvement would be limited to short timescales. Our results support the hypothesis that the cerebellum encodes generalized prediction errors during non-motor learning: We found evidence of reward prediction errors in Crus I/II of the human cerebellum (Trach, Ou, & McDougle, under review), closely aligning with results in model organisms. During SL, Crus I showed increased activation to surprising stimulus transitions (i.e., statistical prediction errors). These signals increased gradually with learning and were weakened with delayed feedback. These results suggest a general function of the cerebellum in predictive processing with domain-general temporal constraints.

Topic Area: PERCEPTION & ACTION: Motor control