Multiclass Sparse Bayesian Regression for fMRI-Based Prediction
Figure 4
Two-dimensional slices of the three-dimensional volume of simulated data. Weights found by different methods, the true target (a) and F-score (b). The Gibbs-MCBR method (d) retrieves almost the whole spatial support for the weights. The sparsity-promoting reference methods, elastic net (f) and ARD (h), find an overly sparse support of the weights. VB-MCBR (c) converges to a local maximum similar to BRR (g) and thus does not yield a sparse solution. SVR (e) yields smooth maps that are not similar to the ground truth.
