A hippocampal population code for rapid generalization

Generalizing from experience and applying prior knowledge to new situations is essential for intelligent behavior. Traditional models attribute this generalization to gradual statistical learning in the neocortex. However, such a slow process cannot account for animals rapid generalization from limited experience. Here, we demonstrate that the hippocampus supports rapid generalization in mice by generating disentangled memory representations, where different aspects of experience are encoded independently. This code enabled the transfer of prior knowledge to solve new tasks. We identify specific circuit mechanisms underlying this rapid generalization. We show that the seemingly random changes in individual neuronal activity over time and across environments result from structured circuit-level processes, governed by the dynamics of local inhibition and cross-regional cell assemblies, respectively. Our findings provide computational and mechanistic insights into how the geometric structure and underlying circuit organization of hippocampal population dynamics facilitate both memory discrimination and generalization, enabling efficient and flexible learning.