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

Citations Over TimeTop 1% of 2015 papers

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 the learnable activation and advanced initialization, 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% [33]). To our knowledge, our result is the first to surpass the reported human-level performance (5.1%, [26]) on this dataset.

Related Papers

• → Automated Brain Image Classification Based on VGG-16 and Transfer Learning(2019)199 cited
• → Image classification with caffe deep learning framework(2017)44 cited
• → Deep and Lightweight Neural Network for Histopathological Image Classification(2022)3 cited
• → Challenges of Deep Learning Methods for COVID-19 Detection Using Public Datasets(2020)10 cited
• → Target Detection and Network Optimization: Deep Learning in Face Expression Feature Recognition(2022)