Title: Distinct Cortical and Subcortical Functions Emerge When Cortical Memory Capacity Is Restricted

Abstract:

The prevailing theory suggests that the brain achieves superior, resource-efficient performance by combining the flexible but computationally demanding processing of the cortex with the simpler, lower-cost mechanisms of subcortical structures. Although this perspective is appealing, comprehensive theoretical models validating this hypothesis remain scarce. To address this gap, we build upon existing frameworks where model-based and model-free modules learn simultaneously, introducing explicit constraints on the memory resources available to the model-based component. We analyze the consequences of these limitations within a basic decision-making context. Our results indicate that memory constraints inherently drive the development of specific strategies for resource allocation. By evaluating the efficacy of various strategies across different scenarios, we find that in environments where rewarded states shift frequently, it is beneficial for the model-based module to prioritize understanding the general structure of the environment over merely exploiting immediate rewards. This study establishes a theoretical basis for the functional separation of cortical and subcortical systems during the learning process: it posits that the cortex is responsible for learning general environmental structures, whereas subcortical circuits specialize in reward-based learning. Additionally, we outline methods for testing these hypotheses using experimental data.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC