Provided by

arXiv

arxiv.org
𝕏gh
0.0/5
5 views
0 saved
Freeadvanced
ResearchersAI EngineersData ScientistsStudents
Introduced AlexNet, a pioneering deep convolutional neural network, achieving breakthrough ImageNet classification performance and igniting the dee...

Overview

This landmark paper introduced AlexNet, a deep convolutional neural network that shattered previous records in the ImageNet LSVRC-2010 competition. It demonstrated the unprecedented power of deep learning for computer vision, utilizing novel techniques like ReLU activations, dropout regularization, and efficient GPU implementation for training its 60-million-parameter, eight-layer architecture. AlexNet's success drastically shifted the research paradigm, making deep CNNs the dominant approach for image classification and setting the stage for the modern AI era.

Abstract

We train a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into 1000 different classes. On the test data, we achieved record-breaking results. The network contains 60 million parameters and 650,000 neurons, and consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overfitting in the fully-connected layers, we employed a newly developed regularization method called “dropout” that proved to be very effective. We also used a new data augmentation scheme to increase the size of the training set. We achieved a top-1 error rate of 37.5% and a top-5 error rate of 17.0% on the ImageNet LSVRC-2010 test set, which was significantly better than the previous state-of-the-art. The network’s depth is essential for its high performance, which is computationally expensive to train, but feasible thanks to fast GPUs. We believe that such large and deep neural networks are feasible to train on available hardware and can achieve state-of-the-art results on challenging image classification tasks.