The cerebellum is an integral part of the human brain and understanding its role in learning might present an opportunity for reciprocal enrichment of the fields of artificial intelligence and neuroscience. In this paper, we present a novel idea that the cerebellum's role goes beyond functioning as a supervised learning machine to performing model-based reinforcement learning. We revisit the current theories about the cerebellum's role in human learning processes and propose a novel way of evaluating it in the context of the simulated cerebellum. Based on the recent experimental findings, we propose that the cerebellum performs model-based reinforcement learning and we propose a way to evaluate the hypothesis using a simulated cerebellum. Finally, we discuss the necessary conditions to evaluate the hypothesis and the potential implications for future research of the hypothesis holds.

PDF

In Adaptive and Learning Agents Workshop at AAMAS, Virtual, May 2021.

@inproceedings{ALA2021-BHARATH, title={Is the Cerebellum a Model-Based Reinforcement Learning Agent?}, author={Bharath Masetty and Reuth Mirsky and Ashish D. Deshpande and Michael Mauk and Peter Stone}, booktitle={Adaptive and Learning Agents Workshop at AAMAS}, month={May}, address={Virtual}, url="http://nn.cs.utexas.edu/?ALA2021-BHARATH", year={2021} }

Slides (PDF)