Leakage and Second-Order Dynamics Improve Hippocampal RNN Replay
2026-02-20 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors studied how noisy recurrent neural networks (RNNs), which mimic brain activity during memory replay, generate their patterns over time. They found that the way replay activity evolves is complex and suggested using a method called hidden state leakage to better model replay. They also showed that adjusting the network's hidden state encourages exploring new patterns but makes replay slower. To fix this, the authors introduced hidden state momentum, which speeds up replay without losing the ability to explore. Their ideas were tested on simulated navigation tasks and synthetic brain cell data.
recurrent neural networksLangevin samplinghidden state leakagehidden state adaptationhidden state momentumpath integrationneural replayMarkov samplinghippocampusplace cells
Authors
Josue Casco-Rodriguez, Nanda H. Krishna, Richard G. Baraniuk
Abstract
Biological neural networks (like the hippocampus) can internally generate "replay" resembling stimulus-driven activity. Recent computational models of replay use noisy recurrent neural networks (RNNs) trained to path-integrate. Replay in these networks has been described as Langevin sampling, but new modifiers of noisy RNN replay have surpassed this description. We re-examine noisy RNN replay as sampling to understand or improve it in three ways: (1) Under simple assumptions, we prove that the gradients replay activity should follow are time-varying and difficult to estimate, but readily motivate the use of hidden state leakage in RNNs for replay. (2) We confirm that hidden state adaptation (negative feedback) encourages exploration in replay, but show that it incurs non-Markov sampling that also slows replay. (3) We propose the first model of temporally compressed replay in noisy path-integrating RNNs through hidden state momentum, connect it to underdamped Langevin sampling, and show that, together with adaptation, it counters slowness while maintaining exploration. We verify our findings via path-integration of 2D triangular and T-maze paths and of high-dimensional paths of synthetic rat place cell activity.