Deep Learning Quick Reference

Overview of transfer learning

In Chapter 7, Convolutional Neural Networks, we trained a convolutional neural network on about 50,000 observations and we saw that, because of the complexity of the network and problem, we were overfitting on the training set after just a few epochs. If you recall, I had made the comment that 50,000 observations in our training set wasn't very large for a computer vision problem. That's true. Computer vision problems love data and the more data we can give them, the better they perform.

The deep neural networks that we might consider state-of-the-art in computer vision are often trained on a dataset called ImageNet. The ImageNet dataset (http://www.image-net.org/) is a 1,000 class classifier that contains 1.2 ...