Multiclass Sparse Bayesian Regression for fMRI-Based Prediction

Algorithm 1

VB-MCBR algorithm.

Initialize 𝑎₁=𝛼₁, 𝑎₂=𝛼₂, 𝑙₁=𝜆₁, 𝑙₂=𝜆₂ and 𝑑_𝑘=𝜂_𝑘

Randomly initialize 𝑞(𝑧_𝑗=𝑘)

Set a number of iterations max steps

repeat

 Compute 𝐴 using (A.1), Σ using (A.2) and 𝜇 using (A.3).

 Compute 𝑙₁ using (A.4) and 𝑙₂ using (A.5).

 Compute 𝑎₁ using (A.6) and 𝑎₂ using (A.7).

 Compute 𝜌_𝑗𝑘 using (A.8).

 Compute 𝜋_𝑘 using (A.9) and 𝑑_𝑘 using (A.10).

until max steps;

Return  𝜇.