AlexNet brings the pieces together

In 2012, a research team entered a deep neural network in the ImageNet challenge. The system became known as AlexNet.

AlexNet brought together three things this module has followed.

A trainable deep network. It had 8 learned layers, about 60 million adjustable weights, and training improvements including ReLU.

Enough computing power. Two graphics cards performed the heavy arithmetic. Training took about five to six days.

Enough data and a fair test. Those 1.2 million labelled images were what it learned from, and it then faced the same hidden test as every other entry.

No single ingredient explains what happened next. The result came from making these pieces work together.