A temporal context model of spatial memory

Decades of research have shown that temporal encoding context is a powerful organizing principle in memory. When remembering lists of words, subjects tend to recall items together that were presented in nearby list positions (“temporal clustering”). The retrieved context framework explains these results by assuming that new memories are stored along with a slowly changing context representation, which cues similar memories at retrieval. Outside of the laboratory, our experiences unfold over both time and space, raising the question of how spatial context affects memory organization. Studies using virtual environments to simulate a varying spatial encoding context, show that spatial distances likewise organize subjects’ memories, such that subjects recall items together that were encoded in nearby spatial locations. Unlike temporal recall organization, this “spatial clustering” effect depends on a learned cognitive map of an environment’s spatial layout. The hippocampus has been implicated in both spatial navigation and episodic memory, and the discovery of hippocampal place cells, time cells and concept cells has motivated an ongoing effort to establish an overarching theory of MTL function. Here, we present a formal computational model that (1) learns spatial maps from a slowly changing temporal context vector and (2) stores and retrieves individual items along with their spatio-temporal context. By running simulations, we show that our model acquires spatial knowledge over time, resulting in both spatial and temporal clustering effects during episodic recall. As such, our model provides a parsimonious, unified account of cognitive mapping and episodic memory in medial temporal lobe circuits.