Across all voxel selections, PCA always produced lower accuracies than VAE–GAN—though this difference did not reach statistical significance given our limited subject number (across all three voxel selections, χ 2 (1) ≥ 3, p > 0.08).
← all excerpts
Reconstructing faces from fMRI patterns using deep generative neural networks.
1
0.0800
0.0800