Summary

We started this chapter by learning about how an image is represented. Then, we learned about how scaling, the value of the learning rate, our choice of optimizer, and the batch size help improve the accuracy and speed of training. We then learned about how batch normalization helps to increase the speed of training and addresses the issues of very small or large values in a hidden layer. Then, we learned about scheduling the learning rate to increase accuracy further. We then proceeded to understand the concept of overfitting and learned about how dropout and L1 and L2 regularization help us avoid overfitting.

Now that we have learned about image classification using a deep neural network, as well as the various hyperparameters that help train a model, in the next chapter, we will learn about how what we’ve learned in this chapter can fail and how to address this, using convolutional neural networks.