In this section, we will train an MLP to recognize facial emotions in the pictures.
We have previously made the point that finding the features that best describe the data is often an essential part of the entire learning task. We have also looked at common preprocessing methods, such as mean subtraction and normalization.
Here, we will look at an additional method that has a long tradition in face recognition—that is, PCA. We are hoping that, even if we don't collect thousands of training pictures, PCA will help us get good results.
