Cognitive Constraints and Reward Environments Jointly Shape Memory Formation

A key adaptive feature of human episodic memory is its ability to selectively encode important information. While extensive empirical evidence supports that reward enhances memory, this monotonic relationship does not always hold, and the influence of reward on memory formation can exhibit complex patterns. To reconcile these findings, we present a computational model that considers not only how people maximize their rewards but also the cognitive constraints underlying this adaptation. Unlike previous theoretical accounts that assume a direct link between reward value and memory encoding strength, our model adaptively decides how strongly to encode each item based on the overall reward environment and its interaction with limited cognitive resources at encoding. We validated the model's predictions across three experiments, successfully explaining why high-reward items do not always have a memory advantage, how reward context influences memory, and the role of reward anticipation. Importantly, the model predictions are parameter-free. They are derived solely from optimal adaptation to a given reward environment, under cognitive constraints independently characterized by previous studies, rather than being fit to the empirical data that the model seeks to explain. These findings support that memory encoding is an active process involving meta-cognitive control, where limited cognitive resources are strategically allocated to maximize overall cumulative rewards, rather than a passive response to the salience of individual items.