Room 4 · The Awakening · 2012
AlexNet Wins ImageNet
AlexNet Wins ImageNet (Krizhevsky, Sutskever and Hinton, 2012): The year deep learning moved to the center of the field.
In 2012, an eight-layer convolutional network with about 60 million parameters won the ImageNet image recognition contest with a 15.3% error, against 26.2% for the runner-up. It was trained on two ordinary graphics cards, the kind used for video games.
To understand AlexNet you first need ImageNet. Starting in 2007, a team led by Fei-Fei Li, then at Princeton and later at Stanford, built a database of millions of photos labeled by people, sorted into thousands of categories. From 2010 it ran an annual contest, the ImageNet Large Scale Visual Recognition Challenge, with about 1.2 million training images in 1,000 categories, from dog breeds to mushrooms.
In 2012 a team from the University of Toronto entered: Alex Krizhevsky, Ilya Sutskever and their supervisor Geoffrey Hinton. Their convolutional network, later known as AlexNet, had eight learned layers and about 60 million parameters. In the contest's main measure, the share of images where the correct label was not among the network's top five guesses, AlexNet scored 15.3 percent. The next best entry scored 26.2 percent.
The ingredients were not new one by one: convolutional networks went back to LeCun's ZIP codes in Room 3, and backpropagation to 1986. What was new was the scale and a few practical tricks. The network was trained on two consumer graphics cards, GPUs designed for video games, which can do many simple calculations in parallel. It used a simpler activation function, ReLU, that sped up training, and a technique called dropout that randomly switches off neurons during training to prevent memorization.
The result convinced much of the field almost overnight. Within a few years nearly every top entry in the contest used deep neural networks, and companies began competing to hire the researchers. Google acquired the team's small company in 2013.
The paper, "ImageNet Classification with Deep Convolutional Neural Networks," was presented at the NIPS conference, now NeurIPS, in December 2012.
Ask your AI: why are graphics cards so good at training neural networks? Ask it to explain without the word "matrix" first.
Source: papers.nips.cc/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstra
See it in the museum →Cite this: Nazareno, I. (2026). AlexNet Wins ImageNet (2012). New Mind Museum. Retrieved November 17, 2026, from https://newmindmuseum.com/museum/alexnet-wins-imagenet-2012/