Summary
The 2012 ImageNet competition. AlexNet (Krizhevsky, Sutskever, Hinton) beat the second-place entry by more than ten percentage points. The result was so decisive that it ended the debate about whether deep learning worked. Every major research group pivoted within a year.
Objectives
- 01State the 2012 top-5 error gap.
- 02Name AlexNet's three key ingredients.
- 03Explain what shipped from ImageNet into industry.
The Lesson
The result
AlexNet: 15.3% top-5 error. Second place: 26.2%. A gap of that size on a well-studied benchmark forces the field to update.
The recipe
Deep CNN, ReLU activations, dropout regularization, trained on two GPUs. Every subsequent vision system descends from that stack — until vision transformers began displacing convolutions after 2020.
Key Ideas
- ImageNet 2012 is the moment deep learning became the default.
- The stack — CNN + ReLU + GPU + large labeled data — was the enabler.
References
- Krizhevsky, Sutskever, Hinton, 'ImageNet Classification with Deep Convolutional Neural Networks,' NIPS (2012)