Machine Learning in Elixir

Wrapping Up

In this chapter, you implemented a convolutional neural network (CNN) and compared its performance to a traditional MLP on computer vision tasks. You broke down the convolution operation, convolutional layers, and max pooling layers. You also learned why convolutional layers are able to learn to represent images so well. Finally, you used a few model training tricks in Axon to improve the performance of your model.

One thing you might have noticed throughout this chapter is that training neural networks can sometimes be tedious and time-consuming. Pushing the performance of your model into a state of the art is no joke. Wouldn’t it be nice if you could save some training time and make use of someone else’s hard work? Fortunately, ...