Multiclass Sparse Bayesian Regression for fMRI-Based Prediction

Figure 1

Graphical model of Multiclass Sparse Bayesian Regression (MCBR). We denote by 𝐲∈ℝ^𝑛 the targets to be predicted and by 𝐗∈ℝ^𝑛×𝑝 the set of activation images. both the weights of the model 𝐰 depend on a discrete variable 𝐳 that assigns each feature to a class 𝑘 among 𝐾. Both the noise 𝜖 and the weights 𝐰 have a Gamma prior on their precisions. The variable 𝐳 follows a Dirichlet prior 𝜋.