Why ReLU Units Sometimes Die: Analysis of Single-Unit Error Backpropagation in Neural Networks
This addresses a practical training issue for neural network practitioners, but it is incremental as it builds on existing understanding of ReLU behavior.
The paper investigates the phenomenon of 'dying ReLU units' in neural networks, showing through simulations on CIFAR-10 that activation probabilities drop below 0.5 in layers without skip connections and decrease towards the output, and provides a statistical analysis indicating slower convergence regardless of weight initialization.
Recently, neural networks in machine learning use rectified linear units (ReLUs) in early processing layers for better performance. Training these structures sometimes results in "dying ReLU units" with near-zero outputs. We first explore this condition via simulation using the CIFAR-10 dataset and variants of two popular convolutive neural network architectures. Our explorations show that the output activation probability Pr[y>0] is generally less than 0.5 at system convergence for layers that do not employ skip connections, and this activation probability tends to decrease as one progresses from input layer to output layer. Employing a simplified model of a single ReLU unit trained by a variant of error backpropagation, we then perform a statistical convergence analysis to explore the model's evolutionary behavior. Our analysis describes the potentially-slower convergence speeds of dying ReLU units, and this issue can occur regardless of how the weights are initialized.