A Unified Cortical Circuit Model with Divisive Normalization and Self-Excitation for Robust Representation and Memory Maintenance

ICLR 2026 Conference Submission18397 Authors

Keywords: Divisive Normalization, Continuous Attractor, Neural Circuit Model, Working Memory, Noise Suppression

Abstract: Robust information representation and its persistent maintenance are fundamental for higher cognitive functions. Existing models employ distinct neural mechanisms to separately address noise-resistant processing or information maintenance, yet a unified framework integrating both operations remains elusive---a critical gap in understanding cortical computation. Here, we introduce a recurrent neural circuit model that synergistically integrates divisive normalization with self-excitation, achieving robust encoding and stable maintenance of normalized input signals. Mathematical analysis reveals that the system exhibits continuous attractor dynamics under appropriate parameters, with two hallmark computational properties: (1) Input-proportional stabilization during stimulus presentation, and (2) self-sustained memory states persisting beyond stimulus offset. The model's functional versatility is evidenced through two principled demonstrations: a) Noise-robust neural encoding in random-dot kinematogram (RDK) task, and b) Approximate Bayesian belief updating in probabilistic Wisconsin Card Sorting Test (pWCST). Our work establishes a unified mathematical framework bridging noise suppression, working memory, and approximate Bayesian inference using a single cortical microcircuit model, offering new perspectives for understanding the brain's canonical computation strategies and developing biologically-plausible artificial neural architectures.

Primary Area: learning on time series and dynamical systems

Submission Number: 18397