Empirical Evaluation of Rectified Activations in Convolutional Network

TLDR

The study investigates how different rectified activation functions—ReLU, Leaky ReLU, PReLU, and a novel randomized Leaky ReLU—affect convolutional neural network performance. The authors evaluate these activations on standard image‑classification benchmarks, primarily the CIFAR‑100 dataset. Results show that adding a non‑zero negative slope consistently improves accuracy, challenging the notion that ReLU sparsity drives performance; deterministic negative slopes overfit on small datasets, whereas the randomized variant achieves 75.68 % accuracy on CIFAR‑100 without ensembles.

Abstract

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.