eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
The source code for AlexNet, a key network in the development of the artificial neural nets that enabled modern AI, is now open source. Researchers can explore it at the Computer History Museum’s GitHub page.
The source code for AlexNet is publicly available now in part because Computer History Museum curator Hansen Hsu reached out to its creator Alex Krizhevsky, citing the code’s “historical significance.” Securing the code required five years of negotiation with Google. After all, Google purchased DNNresearch, the company that owned AlexNet. Other repositories of code called AlexNet already existed on the web, but they were recreations of the code based on a 2012 research paper, not the original.
The birth of AlexNet and the rise of neutral networks
AlexNet was a collaboration between Ilya Sutskever and Alex Krizhevsky — then graduate students at the University of Toronto — and Geoffrey Hinton, their faculty advisor. Hinton and his grad students had already been working on using GPUs to train neural networks for both image and speech recognition, but there wasn’t enough data available at the time to be able to use neural networks for image recognition widely.
Meanwhile, Stanford professor Fei-Fei Li was working on a remarkable but niche project called ImageNet, a database of hand-labeled images that could be used to train computer vision networks. In 2010, she ran a competition to improve the detection of objects with computer vision. In 2011, Sutskever suggested to Krizhevsky that he could train a convolutional neural network for the ImageNet challenge. Hinton joined as the principal investigator on the project.
“Ilya thought we should do it, Alex made it work, and I got the Nobel prize,” Hinton told Hsu.
How AlexNet sparked a deep learning revolution
Krizhevsky used NVIDIA’s CUDA code and two NVIDIA GPUs to train AlexNet. He had previously developed cuda-convnet, a convolutional neural network that also used CUDA and GPUs. AlexNet won the ImageNet image recognition contest in 2012.
The paper Krizhevsky presented at a computer vision conference in the fall of 2012 marked a turning point for computer vision. Afterward, most computer vision papers would focus on neural nets.
After winning the ImageNet contest, the three researchers founded a company around ImageNet called DNNResearch, which was later acquired by Google. By 2022, Sutskever was a co-founder of OpenAI and released ChatGPT.
Since then, generative AI has put image recognition in our phones, and NVIDIA has seen a surge in demand for GPUs that can run it. AlexNet’s lineage and progress show how long the history of generative AI really is, even if it took until today to have the processing power and vast amounts of training data for it to work.