Selective Hippocampal Reactivation Driven by Neocortical Feedback Supports Memory Transfer in Predictive-Coding Networks
- 1Universite Paris Cite
Abstract
During sleep and quiet wakefulness, the brain spontaneously revisits neural activity patterns associated with previous experience, a phenomenon known as reactivation, or replay. Reactivation is thought to support systems consolidation, the process by which memory retrieval becomes progressively more dependent on neocortical circuits and less dependent on the hippocampus. Importantly, reactivation is not indiscriminate: some memory traces, such as newly acquired or weakly learned ones, are preferentially reactivated. These observations have motivated the development of computational models accounting for the spontaneous reactivation of memory traces, their prioritisation for replay, and their transfer between the hippocampus and neocortex. In parallel, predictive-coding (PC) models have shown how computations attributed to hippocampal and neocortical circuits can arise from local prediction-error dynamics. Here, we connect these two lines of work by showing how autonomous intrinsic dynamics in coupled PC networks can transfer stored memory content between neural systems. The hippocampus and neocortex are modelled as two coupled PC associative memory networks, with asymmetric coupling motivated by reported interactions between hippocampal area CA1 and the prefrontal cortex. This coupling establishes a closed-loop hippocampal-neocortical dialogue in which hippocampal activity drives neocortical learning, while neocortical feedback biases subsequent hippocampal activity toward content for which neocortical support remains weak. We prove that this feedback steers hippocampal activity toward states of higher neocortical free energy, so that prioritisation emerges from the dialogue itself rather than from an explicit scheduler. In our model, neocortical learning relies exclusively on spontaneous hippocampal activity, without access to the initial training corpus. Nevertheless, we show that the neocortical network acquires correlated memory content faster than direct training on randomly sampled memory patterns. This advantage appeared for sufficiently correlated patterns, grew with their correlation, and persisted when new patterns were incrementally added during transfer. Together, these results provide a PC account in which neocortical feedback organises hippocampal reactivation to support efficient transfer of memory content during systems consolidation.
Related
- Forward reactivation in human hippocampal CA1 supports repetition learning and is preserved in the aging brain
- Hippocampal stimulation timed to memory reactivation shapes human sleep oscillatory dynamics and consolidation
- Phylogeny of Cortical‐Hippocampal Projections Reveals Selective Elimination of Input From Sensory Regions
- Hippocampal CA3 connectomics reveals a gradient of mossy fiber inputs and selective feedforward inhibition onto pyramidal cells
- Predictive and feedback signals differently shape the formation of language representations
BibTeX
@misc{paul2026selective,
title = {{Selective Hippocampal Reactivation Driven by Neocortical Feedback Supports Memory Transfer in Predictive-Coding Networks}},
author = {S. Paul},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.29.755342},
url = {https://www.biorxiv.org/content/10.64898/2026.09.29.755342v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Selective Hippocampal Reactivation Driven by Neocortical Feedback Supports Memory Transfer in Predictive-Coding Networks
AU - S. Paul
PY - 2026
DA - 2026/09/29
PB - bioRxiv
DO - 10.64898/2026.09.29.755342
UR - https://www.biorxiv.org/content/10.64898/2026.09.29.755342v1
AB - During sleep and quiet wakefulness, the brain spontaneously revisits neural activity patterns associated with previous experience, a phenomenon known as reactivation, or replay. Reactivation is thought to support systems consolidation, the process by which memory retrieval becomes progressively more dependent on neocortical circuits and less dependent on the hippocampus. Importantly, reactivation is not indiscriminate: some memory traces, such as newly acquired or weakly learned ones, are preferentially reactivated. These observations have motivated the development of computational models accounting for the spontaneous reactivation of memory traces, their prioritisation for replay, and their transfer between the hippocampus and neocortex. In parallel, predictive-coding (PC) models have shown how computations attributed to hippocampal and neocortical circuits can arise from local prediction-error dynamics. Here, we connect these two lines of work by showing how autonomous intrinsic dynamics in coupled PC networks can transfer stored memory content between neural systems. The hippocampus and neocortex are modelled as two coupled PC associative memory networks, with asymmetric coupling motivated by reported interactions between hippocampal area CA1 and the prefrontal cortex. This coupling establishes a closed-loop hippocampal-neocortical dialogue in which hippocampal activity drives neocortical learning, while neocortical feedback biases subsequent hippocampal activity toward content for which neocortical support remains weak. We prove that this feedback steers hippocampal activity toward states of higher neocortical free energy, so that prioritisation emerges from the dialogue itself rather than from an explicit scheduler. In our model, neocortical learning relies exclusively on spontaneous hippocampal activity, without access to the initial training corpus. Nevertheless, we show that the neocortical network acquires correlated memory content faster than direct training on randomly sampled memory patterns. This advantage appeared for sufficiently correlated patterns, grew with their correlation, and persisted when new patterns were incrementally added during transfer. Together, these results provide a PC account in which neocortical feedback organises hippocampal reactivation to support efficient transfer of memory content during systems consolidation.
ER -