FriSem

Suhee Cho, research professional in Professor Jay McClelland’s lab, Department of Pscyhology, Stanford University

Title: BTSP—A Fundamental Building Block of the Adaptive Cognitive Map

Abstract: pAnimals can learn new environments with minimal exploration. They quickly associate spatial information with rewards or aversive outcomes and adjust their behavior accordingly. While the hippocampus is thought to encode spatial and other behaviorally relevant information to support this, how it does so with such efficiency remains unclear. Here, we present a computational model showing that behavioral time-scale synaptic plasticity (BTSP), combined with replay, enables the hippocampus to build a predictive map known as the successor representation. In our model, BTSP constructs and dynamically updates this map by integrating spatial input with cue saliency—even after limited experience. This leads to the over-representation of reward locations in the cognitive map and biases behavior toward those rewards. Replay events further extend this bias to unrewarded locations, helping to resolve the credit assignment problem. Importantly, the representation dynamically adjusts when reward locations change, enabling rapid behavioral adaptation. The model also accounts for aversive behavior when negative outcomes are introduced. Together, these findings suggest a biologically plausible mechanism by which BTSP and replay jointly support fast, flexible learning through dynamic updates to the cognitive map.