AlexNet wakes up deep learning

A neural network trained on two gaming GPUs halved the error rate of the best computer-vision systems at ImageNet, stunning the field.

Why it matters

Three cheap ingredients — GPUs, big data, backpropagation — made machine perception work. The template generalize(d) to speech, translation and generation; the AI boom begins at this benchmark.

People

Alex Krizhevsky · Ilya Sutskever · Geoffrey Hinton

Threads

This event sits on Computation.

Related events

• 2017 – 2022

  Attention, transformers, and the language models

  Google's 'Attention Is All You Need' (2017) defined the transformer; GPT-3 (2020) showed scale worked; ChatGPT (2022) put a conversational model in front of a hundred million people in two months.

  Computing

• 1947

  The transistor, three weeks after Christmas

  Bardeen and Brattain demonstrated a point-contact transistor at Bell Labs on December 23, 1947, with Shockley's team driving the solid-state program.

  Computing · americas

• 1945

  ENIAC switches on

  The first general-purpose electronic computer filled a room, weighed 27 tons, and recalculated artillery tables in seconds instead of days.

  Computing · americas

Sources

• Communications of the ACM — ImageNet classification