Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification

TL;DR

This paper introduces PReLU, a parametric rectifier activation function, and a robust initialization method, enabling the training of very deep neural networks that surpass human-level performance on ImageNet classification.

Contribution

The paper presents PReLU and a new initialization technique, allowing training of deeper networks and achieving state-of-the-art results surpassing human performance.

Findings

PReLU improves model fitting with minimal extra cost.

Deep rectified networks can be trained from scratch.

Achieved 4.94% top-5 error on ImageNet, surpassing human performance.

Abstract

Rectified activation units (rectifiers) are essential for state-of-the-art neural networks. In this work, we study rectifier neural networks for image classification from two aspects. First, we propose a Parametric Rectified Linear Unit (PReLU) that generalizes the traditional rectified unit. PReLU improves model fitting with nearly zero extra computational cost and little overfitting risk. Second, we derive a robust initialization method that particularly considers the rectifier nonlinearities. This method enables us to train extremely deep rectified models directly from scratch and to investigate deeper or wider network architectures. Based on our PReLU networks (PReLU-nets), we achieve 4.94% top-5 test error on the ImageNet 2012 classification dataset. This is a 26% relative improvement over the ILSVRC 2014 winner (GoogLeNet, 6.66%). To our knowledge, our result is the first to surpass human-level performance (5.1%, Russakovsky et al.) on this visual recognition challenge.