AlexNet is the deep convolutional neural network that won the 2012 ImageNet image recognition challenge by a wide margin, showing that deep networks trained on raw pixels could beat older methods.
In plain words
Built by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton at the University of Toronto, it learned to sort 1.3 million photos into 1,000 categories straight from raw pixels. It had 60 million parameters, five convolutional layers, then fully connected layers ending in 1,000 outputs. Two tricks made it work: training on GPUs for speed, and a then-new regularisation method to reduce overfitting. Its top-5 error on the 2012 test set was about 15 to 16 per cent, far ahead of the next entry.
A quick example
Imagine showing it a photo of a cat. The convolutional layers turn the pixels into more and more abstract features, and the final layer gives a score for each of the 1,000 classes, with "tabby cat" hopefully the highest.
Where you'll meet it
- The model that showed deep networks beat hand-built image features.
- A classic teaching example, still available in PyTorch.
Related
Sources
- PaperImageNet Classification with Deep Convolutional Neural Networks, Krizhevsky, Sutskever and Hinton (NeurIPS) · read 29 Sept 2026
- DocsILSVRC2012 results, ImageNet · read 29 Sept 2026
- DocsAlexNet, PyTorch torchvision docs · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.