Dynamical Isometry and a Mean Field Theory of CNNs: How to Train 10,000-Layer Vanilla Convolutional Neural Networks

Abstract

In recent years, state-of-the-art methods in computer vision have utilized increasingly deep convolutional neural network architectures (CNNs), with some of the most successful models employing hundreds or even thousands of layers. A variety of pathologies such as vanishing/exploding gradients make training such deep networks challenging. While residual connections and batch normalization do enable training at these depths, it has remained unclear whether such specialized architecture designs are truly necessary to train deep CNNs. In this work, we demonstrate that it is possible to train vanilla CNNs with ten thousand layers or more simply by using an appropriate initialization scheme. We derive this initialization scheme theoretically by developing a mean field theory for signal propagation and by characterizing the conditions for dynamical isometry, the equilibration of singular values of the input-output Jacobian matrix. These conditions require that the convolution operator be an orthogonal transformation in the sense that it is norm-preserving. We present an algorithm for generating such random initial orthogonal convolution kernels and demonstrate empirically that they enable efficient training of extremely deep architectures.

arXiv commentICML 2018 Conference Proceedings

References in corpus (8)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Natural Language Processing (almost) from Scratch
• Exact solutions to the nonlinear dynamics of learning in deep linear neural networks
• Exponential expressivity in deep neural networks through transient chaos
• The Emergence of Spectral Universality in Deep Networks
• Universal Statistics of Fisher Information in Deep Neural Networks: Mean Field Approach
• On the Selection of Initialization and Activation Function for Deep Neural Networks
• A Correspondence Between Random Neural Networks and Statistical Field Theory

Cited by in corpus (7)

• Feedback alignment in deep convolutional networks
• Wide and Deep Neural Networks Achieve Optimality for Classification
• A Signal Propagation Perspective for Pruning Neural Networks at Initialization
• The Nonlinearity Coefficient - Predicting Generalization in Deep Neural Networks
• Feedback Gradient Descent: Efficient and Stable Optimization with Orthogonality for DNNs
• Regularization in ResNet with Stochastic Depth
• A general learning system based on neuron bursting and tonic firing