English

A Mean Field Theory of Batch Normalization

Neural and Evolutionary Computing 2019-03-07 v2 Disordered Systems and Neural Networks Machine Learning Dynamical Systems

Abstract

We develop a mean field theory for batch normalization in fully-connected feedforward neural networks. In so doing, we provide a precise characterization of signal propagation and gradient backpropagation in wide batch-normalized networks at initialization. Our theory shows that gradient signals grow exponentially in depth and that these exploding gradients cannot be eliminated by tuning the initial weight variances or by adjusting the nonlinear activation function. Indeed, batch normalization itself is the cause of gradient explosion. As a result, vanilla batch-normalized networks without skip connections are not trainable at large depths for common initialization schemes, a prediction that we verify with a variety of empirical simulations. While gradient explosion cannot be eliminated, it can be reduced by tuning the network close to the linear regime, which improves the trainability of deep batch-normalized networks without residual connections. Finally, we investigate the learning dynamics of batch-normalized networks and observe that after a single step of optimization the networks achieve a relatively stable equilibrium in which gradients have dramatically smaller dynamic range. Our theory leverages Laplace, Fourier, and Gegenbauer transforms and we derive new identities that may be of independent interest.

Keywords

Cite

@article{arxiv.1902.08129,
  title  = {A Mean Field Theory of Batch Normalization},
  author = {Greg Yang and Jeffrey Pennington and Vinay Rao and Jascha Sohl-Dickstein and Samuel S. Schoenholz},
  journal= {arXiv preprint arXiv:1902.08129},
  year   = {2019}
}

Comments

To appear in ICLR 2019

R2 v1 2026-06-23T07:47:20.984Z