After knowing some basic facts of Deep Learning, the course goes further to how to better the training process.
Since there are plenty of parameters and hyper parameters in a NN, obviously, there are a lot of things we can try to better the model.
Before starting the actual tuning, we should know how to estimate the quality of a NN.
Compared with under fitting, over fitting is a greater problem.
Here are some methods to help with over fitting, including regularization, inverted dropout, and some other ways.
Adding more data into the training set is always a choice, but there are also something you can do to enlarge the set, even without brand new data.
After building a model that predict the outcome to the extent of 'just right', there are some technics which can help with the optimization of the process.
Now that we know what we can do to better the training process, we should take a look at the exsiting deep learning frameworks, and see how to make a wise choice.
