A deep convolutional network wins the ImageNet challenge

Assembled from 36 blocks · 1 evidence · 35 related

1. Story
2. The network entered by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton achieved a top-5 error rate of 15.3% on the ImageNet Large Scale Visual Recognition Challenge against 26.2% for the runner-up — a margin large enough that the method, rather than the increment, became the story. It was an eight-layer convolutional network trained on graphics processors with dropout regularisation, and it learned its own features rather than being given hand-designed ones. Within about two years essentially every serious entry in computer vision was a deep network. This is the conventional starting point for the present era, and it is a defensible one: it is the moment a public, externally scored, held-out test was won by a wide margin by a method the field had largely written off.
3. Knowledge
4. Deep learning
5. ImageNet
6. The AI accelerator
7. Geoffrey Hinton
8. A deep convolutional network wins the ImageNet challenge
9. Connections
10. Geoffrey Hinton
11. Deep learning
12. ImageNet
13. Geoffrey Hinton
14. The AI accelerator
15. The AI accelerator
16. ImageNet
17. The transformer architecture
18. AlphaFold
19. Machine-learned weather forecasting
20. Machine learning in medical imaging
21. A deep convolutional network wins the ImageNet challenge
22. AlphaGo defeats Lee Sedol in Seoul
23. Deep learning
24. Fei-Fei Li
25. ImageNet is assembled and released
26. A deep convolutional network wins the ImageNet challenge
27. Machine learning in medical imaging
28. A deep convolutional network wins the ImageNet challenge
29. A deep convolutional network wins the ImageNet challenge
30. Deep learning
31. The transformer architecture
32. Training compute
33. Taiwan
34. A deep convolutional network wins the ImageNet challenge
35. Evidence
36. An eight-layer convolutional network — five convolutional and three fully connected layers — trained on graphics processors with dropout regularisation achieved a top-5 error rate of 15.3% at ILSVRC 2012 against 26.2% for the second-placed entry, learning hierarchical features rather than using hand-designed ones. V55 verification basis: the paper was not fetched. The 15.3% and 26.2% figures and the architecture description were carried in multiple retrieved summaries and in a hosted copy of the paper itself, but the attribution in those summaries is to secondary explainers as often as to the paper, and a curator should confirm both error rates against the ILSVRC 2012 results table before publication.