Adaptive self-organized criticality in deep neural networks
Abstract
Deep neural networks are high-dimensional dynamical systems whose function depends on the stable propagation of activity and perturbations across many layers. Maintaining suitable dynamical regimes may therefore be essential for robust learning and for preventing dynamical instabilities during training. Here, we show that the global dynamical state of a deep neural network can be autonomously regulated by purely local homeostatic plasticity. Neuronal activity is inferred from responses across inputs, and individual synapses are strengthened or weakened using only the activity of their postsynaptic neuron. Without measuring any global network property, this rule drives networks from both subcritical and supercritical initial conditions toward a common critical state, characterized by conserved activity propagation and a vanishing largest finite-time Lyapunov exponent. When combined with gradient-based learning, homeostatic adaptation counteracts the training-induced drift toward supercritical dynamics, while revealing a competition between dynamical regulation and task optimization. Our results demonstrate how adaptive self-organization can be implemented in deep neural networks and how local plasticity can control their collective dynamical operating point.
Related
- Prospective Coding and Path Integration Emerge as Equilibrium Solutions of Self-Organizing Neural Networks with Firing-Rate Adaptation
- Task-evoked flexibility of intrinsic neural timescales is hierarchically organized, behaviorally relevant, and refined across two developmental stages
- Self-organized hummock-hollow patterns redistribute salinity stress in coastal meadows
- Self-organized Regulation of Group Size and Number in Natural and Artificial Collectives
- Self-organized spectral adjustment in echolocation sounds within group-flying bats.
BibTeX
@misc{vock2026adaptive,
title = {{Adaptive self-organized criticality in deep neural networks}},
author = {Simon Vock and Christian Meisel},
howpublished = {arXiv},
year = {2026},
month = aug,
doi = {10.48550/arxiv.2608.28431},
eprint = {2608.28431},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.28431},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Adaptive self-organized criticality in deep neural networks
AU - Simon Vock
AU - Christian Meisel
PY - 2026
DA - 2026/08/28
PB - arXiv
DO - 10.48550/arxiv.2608.28431
UR - https://arxiv.org/abs/2608.28431
AB - Deep neural networks are high-dimensional dynamical systems whose function depends on the stable propagation of activity and perturbations across many layers. Maintaining suitable dynamical regimes may therefore be essential for robust learning and for preventing dynamical instabilities during training. Here, we show that the global dynamical state of a deep neural network can be autonomously regulated by purely local homeostatic plasticity. Neuronal activity is inferred from responses across inputs, and individual synapses are strengthened or weakened using only the activity of their postsynaptic neuron. Without measuring any global network property, this rule drives networks from both subcritical and supercritical initial conditions toward a common critical state, characterized by conserved activity propagation and a vanishing largest finite-time Lyapunov exponent. When combined with gradient-based learning, homeostatic adaptation counteracts the training-induced drift toward supercritical dynamics, while revealing a competition between dynamical regulation and task optimization. Our results demonstrate how adaptive self-organization can be implemented in deep neural networks and how local plasticity can control their collective dynamical operating point.
ER -