Multiclass Sparse Bayesian Regression for fMRI-Based Prediction
Figure 3
Results on simulated regression data. Weights of the first two features found for the different steps of Gibbs-MCBR (a) and VB-MCBR (b). The red dot represents the ground truth of both weights, and the green dot represents the final state found by the two algorithms. VB-MCBR is stuck in a local maximum, and Gibbs-MCBR finds the correct weights.
