A predictive coding model of hippocampo-neocortical interactions involved in memory replay

ICLR 2026 Conference Submission21446 Authors

Keywords: predictive coding, neocortex, hippocampus, episodic memory, semantic memory, generative replay, neuroscience, generative model

Abstract: The neocortex and the hippocampus are two complementary learning systems which interact during memory construction and consolidation. The hippocampus stores episodic memories coming from the neocortex passing through the entorhinal cortex, and later replays them back to the neocortex to transform them into semantic memory during memory consolidation. It is thought that memory replay is a generative process, involved in imagining, because new episodes can also be generated and instantiated in the neocortex. Here we present a computational model of hippocampal-neocortical interactions based on a predictive coding network with two hidden layers, which are mapped to the visual cortex and the entorhinal cortex. Improving on a previous implementation of this network, our simulations provide a mechanistic account of memory replay in the neocortex.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 21446