Representational drift under spontaneous activity -- self-organized criticality enhances representational reliability

A team of researchers led by Zhuda Yang has published a groundbreaking paper titled 'Representational drift under spontaneous activity -- self-organized criticality enhances representational reliability' on arXiv. The study tackles a fundamental neuroscience paradox: how does the brain maintain reliable representations of the world while its neural circuits are constantly changing through plasticity and spontaneous activity? The researchers discovered that the brain's spontaneous dynamics operate near a 'critical state'—a highly variable, chaotic regime that paradoxically enhances representational reliability by approximately 40% compared to non-critical states.

Using data from mouse visual cortex and computational modeling, the team built an excitation-inhibition network with homeostatic plasticity that self-organizes to this critical state. Their model successfully reproduced both 'representational drift' (the natural shifting of neural representations over time) and 'restricted representational geometry' observed experimentally. They found that the critical state confines synapse weights to a low-variation space, preventing accumulated drift and enhancing cross-session low-dimensional representation. This mechanism explains how the brain maintains consistent perception and behavior despite continuous synaptic rewiring—a phenomenon that has puzzled neuroscientists for decades.

The findings suggest that self-organized criticality isn't just a ubiquitous property of neural systems but serves a crucial functional purpose. By operating near this chaotic edge, the brain balances flexibility with stability, allowing for learning and adaptation while preserving core representations. This research provides a mathematical framework for understanding brain resilience and could inform more robust AI systems that mimic biological learning mechanisms.