Empirical Evaluation of Rectified Activations in Convolutional Network
This work addresses activation function design for image classification, offering incremental improvements over existing methods.
The paper investigated rectified activation functions in convolutional neural networks, finding that non-zero slopes for negative parts improve performance, challenging the belief that sparsity is key, and achieved 75.68% accuracy on CIFAR-100 with a randomized method.
In this paper we investigate the performance of different types of rectified activation functions in convolutional neural network: standard rectified linear unit (ReLU), leaky rectified linear unit (Leaky ReLU), parametric rectified linear unit (PReLU) and a new randomized leaky rectified linear units (RReLU). We evaluate these activation function on standard image classification task. Our experiments suggest that incorporating a non-zero slope for negative part in rectified activation units could consistently improve the results. Thus our findings are negative on the common belief that sparsity is the key of good performance in ReLU. Moreover, on small scale dataset, using deterministic negative slope or learning it are both prone to overfitting. They are not as effective as using their randomized counterpart. By using RReLU, we achieved 75.68\% accuracy on CIFAR-100 test set without multiple test or ensemble.