Wikinventia — Atlas of discoveries and inventions · Digital Age

AlexNet — the takeoff of modern deep learning — Krizhevsky, Sutskever, and Hinton

2012 AD · Transmission: Global
AIInventionNorth American

In September 2012 AlexNet won the ImageNet Large Scale Visual Recognition Challenge with an error margin 10% lower than the second-place classifier — an unprecedented gap in the competition's history. The network used GPUs for training, a deep convolutional architecture, and dropout for regularization. The result was a turning point: it proved to the entire AI community that deep neural networks, trained on large data with parallel hardware, could radically outperform classical approaches. Within three years, practically all computer-vision research had migrated to deep learning. AlexNet did not invent deep learning — that credit belongs to decades of work by Hinton, LeCun, Bengio, and others — but it demonstrated it irrefutably to the world.

InstitutionUniversity of Toronto
Historical regionToronto, Canada
Primary sourceKrizhevsky, A., Sutskever, I. & Hinton, G.E. — "ImageNet Classification with Deep Convolutional Neural Networks" (NeurIPS, 2012)
Secondary sourceLeCun, Y., Bengio, Y. & Hinton, G. — "Deep learning" (Nature, 521, 436–444, 2015)
Original languageEnglish
View this entry in the interactive atlas → View in graph →