| Summary: | <p>The hippocampal formation has been implicated in both learning and generalization. The large variety of neural representations, ranging from grid and border cells to place and objectvector cells, observed in the hippocampus, entorhinal cortex, and surrounding brain regions are believed to form the neural bases for intelligent learning. In conjunction with countless experimental studies, several computational models of the hippocampal formation have been proposed as explanations for how the neural representations can support learning. Most such models, however, exhibit one of two critical flaws: they either fail to generalize and transfer what they have learned from one environment to another or they can only do so after having observed many training environments and become useless in environments with different underlying structure. Needless to say, a unified explanation for how the hippocampal formation can enable generalizable learning remains elusive. Here, I present a more unified model that combines the two aforementioned model types and reciprocally addresses each other’s weaknesses. I demonstrate that this combined model enables faster generalization while maintaining flexibility across different environment or task structures, rendering this model a potentially more complete description of hippocampal learning.</p>
|
|---|