Summary

In previous chapters, we considered that we know the structure of the network, which is not true in most of real-life cases. In such cases, we need to learn the structures from the data. In this chapter, we discussed the problem of learning the parameters and structures using just data samples. Firstly, we discussed two different techniques of parameter estimation, maximum likelihood estimation, and Bayesian estimation. We saw that in cases when the data samples given to us don't represent the underlying distribution, the Maximum Likelihood estimate fails to generalize over new data points. Then, we discussed the problem of learning the structure from the data using the same two techniques, that is, maximum likelihood and Bayesian learning. We showed that in the case of structure learning as well, maximum likelihood overfits the training data if we don't have enough samples.

In the next chapter, we will discuss the parameters and structures of Markov networks using data samples.