Universal Scaling Laws of Absorbing Phase Transitions in Artificial Deep Neural Networks

Abstract

We demonstrate that conventional artificial deep neural networks operating near the phase boundary of the signal propagation dynamics, also known as the edge of chaos, exhibit universal scaling laws of absorbing phase transitions in non-equilibrium statistical mechanics. We exploit the fully deterministic nature of the propagation dynamics to elucidate an analogy between a signal collapse in the neural networks and an absorbing state (a state that the system can enter but cannot escape from). Our numerical results indicate that the multilayer perceptrons and the convolutional neural networks belong to the mean-field and the directed percolation universality classes, respectively. Also, the finite-size scaling is successfully applied, suggesting a potential connection to the depth-width trade-off in deep learning. Furthermore, our analysis of the training dynamics under the gradient descent reveals that hyperparameter tuning to the phase boundary is necessary but insufficient for achieving optimal generalization in deep networks. Remarkably, nonuniversal metric factors associated with the scaling laws are shown to play a significant role in concretizing the above observations. These findings highlight the usefulness of the notion of criticality for analyzing the behavior of artificial deep neural networks and offer new insights toward a unified understanding of the essential relationship between criticality and intelligence.

arXiv comment15 pages, 5 figures; added ReLU finite-size scaling results, revised texts for clarity

References in corpus (17)

• Nonequilibrium Critical Phenomena and Phase Transitions into Absorbing States
• Colloquium: Criticality and dynamical scaling in living systems
• Dying ReLU and Initialization: Theory and Numerical Examples
• Collapse of Deep and Narrow Neural Nets
• The Principles of Deep Learning Theory
• Self-organized criticality as an absorbing-state phase transition
• Low-density series expansions for directed percolation I: A new efficient algorithm with applications to the square lattice
• Epidemic analysis of the second-order transition in the Ziff-Gulari-Barshad surface-reaction model
• The thermodynamics of computational copying in biochemical systems
• Thermodynamics of accuracy in kinetic proofreading: Dissipation and efficiency trade-offs
• High-precision Monte Carlo study of directed percolation in (d+1) dimensions
• Monte Carlo simulations and numerical solutions of short-time critical dynamics
• Extended Symbolic Dynamics in Bistable CML: Existence and Stability of Fronts
• Entropy Production of Nonequilibrium Steady States with Irreversible Transitions
• A devil's advocate view on 'self-organized' brain criticality
• Suppression of chaos in a partially driven recurrent neural network
• Entropy Governed by the Absorbing State of Directed Percolation