Beyond gist and detail: the hippocampus as a multiscale integrator

Here I advance an account of hippocampal memory grounded in three features: locally integrated computation; representational plasticity; and graded long-axis receptive field sizes. Together, these premises reposition the hippocampus as a multiscale integrator, in which variable tolerances for pattern overlap are fundamental to operation. Lower tolerances (associated with the posterior hippocampus; pHPC) elicit frequent “trace breaks” in memory integration, yielding temporally dense and structured representations with high retroactive interference. These traces comprise an episodic memory store of detailed, sequential, and infrequently accessed information. Higher tolerances (associated with anterior hippocampus; aHPC) elicit less frequent trace breaks, yielding a lower density of sparse, accessible representations that support an atemporal and non-episodic hippocampal memory store of information integrated over overlapping events. By layering these experiences using non-monotonic learning rules, this store establishes summaries and inference without executive overhead, while also supporting episodic search and prospective thinking. Because these memory systems emerge from divergent tuning of a common hippocampal architecture, they share features such as shared circuitry, rapid encoding, and cortical reinstatement. I describe this more integrative aHPC store as tableau memory, based on its temporally folded synthesis of events, and propose testable predictions in the domains of list-learning, novelty, and contextual processing. By abstracting experiences persistently and automatically, this store helps to reconcile the hippocampus’ known roles in memory specificity and integration, and to support constructive operations required for numerous non-memory functions. Overall, the model reframes hippocampal memory as a system that captures and mobilizes meaning across a variety of biologically relevant timescales.