This might be very broad question. Since ML application in Matter Modelling is emerging field it would good to understand how it is applied and why it is useful.
The term "Machine learning" is quite general. Let's look at the three main domains of it :
Supervised-Learning: In this method, you would need to train a neural network or a statistical learning algorithm on a labelled training set. Some examples of its application in DFT involve :
Unsupervised-Learning: This category mainly involves clustering, dimensionality reduction..etc. And you would be training using unlabeled data. Such methods are used for finding out hidden relations that may pre-exist in your data or may be brought about via a basis transformation. The set of techniques falling under this catogory are used mainly for the analysis of the data obtained via DFT. Some examples include :
Reinforcement Learning: In this method, you would find that the learning process depends on a scalar reward. You can imagine it being similar to the Monte-Carlo method with an associated reward. It does share a lot of statistical concepts with MC like markov states. This method is existant in literature but I'm not entirely familiar with its application to DFT.
You can also check this review paper out too.
Hope it helps :)