Fig. 2.

Step 1: Prior to dimensionality estimation, raw data are pre-processed with preferred settings and software and beta estimates derived from a GLM are obtained for each condition of interest. The resulting j matrices of size n (number of voxels) $\times m$ (number of conditions) are pre-whitened and mean-centered (by row, i.e., voxel) to remove baseline differences across runs. Step 2: a combination of cross-validation and SVD is implemented to find the best dimensionality estimate k for each run j.