Xiao et al. show efficient coding under constraints drives brain criticality
New theory links efficiency, criticality, and sloppiness in neural populations.
A new theoretical paper from researchers He Xiao, Xinyue Zhao, and Weikang Wang (arXiv:2605.22598) offers a compelling explanation for why the brain operates near a critical state. Using a Gaussian population coding model, they demonstrate that maximizing Fisher information—a measure of how accurately neural activity encodes stimuli—under resource constraints naturally leads to hallmark features of criticality: soft modes, diverging correlation lengths, and critical slowing down. This provides a functional rationale for neural avalanches that follow power-law distributions, long observed in experiments but poorly understood from first principles.
The framework also accounts for "sloppiness," the widespread observation that neural systems are insensitive to many parameter changes, by showing it emerges from the same optimization that drives criticality. By incorporating spatial structure, the authors unify statistical and dynamical perspectives of criticality, linking diverging correlations with bifurcation dynamics. Numerical simulations confirm that optimization yields power-law responses, tying together efficient coding, sloppiness, and the critical brain hypothesis into a single mechanistic theory.
- Maximizing Fisher information under resource constraints yields diverging correlation lengths and power-law neural avalanches.
- The framework unifies statistical criticality (diverging correlations) with dynamical criticality (critical slowing down and bifurcations).
- Sloppiness—insensitivity to parameter changes—emerges naturally as a byproduct of efficient coding optimization.
Why It Matters
Provides a principled reason for brain criticality, linking coding efficiency to network dynamics—core for AI and neuroscience.