Nat Comm '21 | Clone-structured graph representations enable flexible learning and vicarious evaluation of cognitive maps.

[NorbertZheng]

George D, et. al. Clone-structured graph representations enable flexible learning and vicarious evaluation of cognitive maps.

[NorbertZheng]

Abstract

To form abstract maps, the hippocampus has to learn to separate or merge aliased observations appropriately in different contexts in a manner that enables generalization and efficient planning. In this paper, we propose a specific higher-order graph structure, CSCG, which forms clones of observation for different contexts as a representation that address these problems.

CSCGs can be learned efficiently using a probabilistic sequence model that is inherently robust to uncertainty. By lifting aliased observations into a hidden space, CSCGS reveal latent modularity useful for hierarchical abstraction and planning.