Oscillatory Phase Coding Emerges in Adaptive RNNS for Flexible Working Memory

Oscillatory phase coding emerges spontaneously in adaptive recurrent neural networks trained on sequence working memory with variable delays, with item identity encoded by oscillation amplitude and serial order by phase.

• Why it matters: The computational role and neural substrates of oscillatory phase coding in memory and sequence processing remain unclear, limiting mechanistic understanding of how such dynamics support cognition.
• What they did: The authors trained adaptive RNNs on sequence working memory tasks with variable delays, then used ablations and neuron-type and connectivity analyses to identify requirements and circuit structure.
• The result: Ablations showed OPC needs adaptation in neural dynamics, variable delay during retention, and flexible readout; analyses revealed content and time cells with phase-organized time-cell connections, enabling a circuit model that reproduces OPC.

AI summary of the abstract · may contain errors · not medical advice · read the paper.