ImageNet Classification with Deep Convolutional Neural Networks
In the 1980s backpropagation did not live up to the very high expectations of its advocates. Twenty years later, we know what went wrong: for deep neural networks to shine, they need far more labeled data and hugely more computation.
