Deep neural networks train better with adjusted edge of chaos settings

Finite-width shift of the edge of chaos in deep neural networks

Machine Learning

Summary

Neural networks learn best when their initial settings are balanced so that small changes in input neither fade away nor explode as they pass through layers. Previous knowledge about this balance was based on infinitely wide networks, but real networks have limited width, which shifts this balance point. The authors found a precise way to calculate this shift, improving how networks are initialized for learning, especially in deeper networks. Their method, called Crinit, helps networks learn faster and more accurately without trial-and-error tuning.

What this means in practice

  • For deep learning engineers: Set initial weight and bias variances precisely to improve training speed and accuracy in deep neural networks without extra tuning.
  • For machine learning platform developers: Integrate a theoretical formula to automatically initialize network parameters at the finite-width edge of chaos, reducing failed training attempts in deep models.

Tested on one dataset.

Authors

Arn Caesar

Abstract

Deep neural networks train best when they are initialised at the edge of chaos, the critical line in the plane of weight and bias variances on which the distance between two inputs neither grows nor shrinks with depth. The line is known for infinitely wide networks; at finite width it moves, and how it moves has been disputed. We treat the logarithm of the distance between two inputs through a network of width $N$ as a random walk and compute its drift and variance to order $1/N$. The edge of chaos moves to a larger weight variance, by $Δσ_w^2=-a_1/(N\,\partial_{σ_w^2}\lnχ_1)$, about $8/N$ for $\tanh$ networks at a bias variance of $0.05$, confirmed by simulations of the exact finite-width recursion. The shift describes a typical network; the average over many networks is dominated by rare networks in which the distance grows, and its critical point moves the other way, which explains why calculations and simulations have disagreed. Backpropagation follows the same random walk, so the shift applies to gradients unchanged. Along the edge of chaos, the information about the similarity of two inputs is constant until the compression over the depth exceeds about one nat, which fixes where to sit on the line. The results give an initialisation rule, Crinit, that places a network on its finite-width edge of chaos without a search. On MNIST and Fashion-MNIST at width $64$ it is more accurate than the standard rules at their usual settings, by a small margin in shallow networks and a large one in deep networks, where most standard rules fail; it trains in fewer steps, and it matches or beats the standard rules with a tuned gain.