Machine Learning

13.6 Sparse Bayesian Learning (SBL)

In Section 13.3, the prior for each one of the unknown parameters, θ_k,k = 0,1,…,K − 1, were given the liberty to have their own variances, ${\sigma }_k^2:=\frac{1}{{\alpha }_k}$. In turn, these variances were treated as hidden random variables and a prior was assigned to each of them in terms of a number of hyperparameters.

In [75, 81], the model was slightly modified. The concept of using different variances for the priors was retained, but the variances were treated as deterministic parameters and not as random ones.² In this context, the task becomes a generalization of the one treated in Section 12.6, and it is built upon the following ...