On the Impact of the Activation Function on Deep Neural Networks Training

Abstract

The weight initialization and the activation function of deep neural networks have a crucial impact on the performance of the training procedure. An inappropriate selection can lead to the loss of information of the input during forward propagation and the exponential vanishing/exploding of gradients during back-propagation. Understanding the theoretical properties of untrained random networks is key to identifying which deep networks may be trained successfully as recently demonstrated by Samuel et al (2017) who showed that for deep feedforward neural networks only a specific choice of hyperparameters known as the `Edge of Chaos' can lead to good performance. While the work by Samuel et al (2017) discuss trainability issues, we focus here on training acceleration and overall performance. We give a comprehensive theoretical analysis of the Edge of Chaos and show that we can indeed tune the initialization parameters and the activation function in order to accelerate the training and improve the performance.

arXiv comment22 pages

Cited by in corpus (16)

• Drawing Inspiration from Biological Dendrites to Empower Artificial Neural Networks
• Finite volume method network for acceleration of unsteady computational fluid dynamics: non-reacting and reacting flows
• Sub-Weibull distributions: generalizing sub-Gaussian and sub-Exponential properties to heavier-tailed distributions
• Wide and Deep Neural Networks Achieve Optimality for Classification
• Robust Pruning at Initialization
• Stable behaviour of infinitely wide deep neural networks
• Neuron Campaign for Initialization Guided by Information Bottleneck Theory
• Critical initialisation in continuous approximations of binary neural networks
• Bayesian neural network unit priors and generalized Weibull-tail property
• Wide stochastic networks: Gaussian limit and PAC-Bayesian training
• Initializing ReLU networks in an expressive subspace of weights
• Posterior contraction for deep Gaussian process priors
• Using activation histograms to bound the number of affine regions in ReLU feed-forward neural networks
• InfoNEAT: Information Theory-based NeuroEvolution of Augmenting Topologies for Side-channel Analysis
• Regularization in ResNet with Stochastic Depth
• Guiding Neural Network Initialization via Marginal Likelihood Maximization