On the Selection of Initialization and Activation Function for Deep Neural Networks

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 Schoenholz 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. We complete this analysis by providing quantitative results showing that, for a class of ReLU-like activation functions, the information propagates indeed deeper for an initialization at the edge of chaos. By further extending this analysis, we identify a class of activation functions that improve the information propagation over ReLU-like functions. This class includes the Swish activation, ${\varphi }_{swish}(x)=x\cdot \text{sigmoid}(x)$, used in Hendrycks & Gimpel (2016), Elfwing et al. (2017) and Ramachandran et al. (2017). This provides a theoretical grounding for the excellent empirical performance of ${\varphi }_{swish}$ observed in these contributions. We complement those previous results by illustrating the benefit of using a random initialization on the edge of chaos in this context.

arXiv comment8 pages, 15 figures

References in corpus (7)

• On the Number of Linear Regions of Deep Neural Networks
• Probabilistic Backpropagation for Scalable Learning of Bayesian Neural Networks
• Deep Neural Networks as Gaussian Processes
• Self-Normalizing Neural Networks
• Exponential expressivity in deep neural networks through transient chaos
• Sigmoid-Weighted Linear Units for Neural Network Function Approximation in Reinforcement Learning
• Deep Information Propagation

Cited by in corpus (23)

• Deep Learning vs. Traditional Computer Vision
• Neural Tangents: Fast and Easy Infinite Neural Networks in Python
• Tensor Programs II: Neural Tangent Kernel for Any Architecture
• Finite Versus Infinite Neural Networks: an Empirical Study
• Dynamical Isometry and a Mean Field Theory of CNNs: How to Train 10,000-Layer Vanilla Convolutional Neural Networks
• A Fine-Grained Spectral Perspective on Neural Networks
• How Good is the Bayes Posterior in Deep Neural Networks Really?
• Feature Learning in Infinite-Width Neural Networks
• Parameter Estimation with Dense and Convolutional Neural Networks Applied to the FitzHugh-Nagumo ODE
• Tensor Programs III: Neural Matrix Laws
• Towards NNGP-guided Neural Architecture Search
• Eigenvalue distribution of nonlinear models of random matrices
• Using a thousand optimization tasks to learn hyperparameter search strategies
• Gradient target propagation
• Model Complexity of Deep Learning: A Survey
• Measuring Model Complexity of Neural Networks with Curve Activation Functions
• Analyzing Finite Neural Networks: Can We Trust Neural Tangent Kernel Theory?
• Initializing ReLU networks in an expressive subspace of weights
• Fractional moment-preserving initialization schemes for training deep neural networks
• Sample Variance Decay in Randomly Initialized ReLU Networks
• Frequency Disentangled Residual Network
• Mean Field Theory of Activation Functions in Deep Neural Networks
• The SWAG Algorithm; a Mathematical Approach that Outperforms Traditional Deep Learning. Theory and Implementation