Fig. 2.

Feature selection results. To train and test the neural action prediction model, we selected voxels sensitive to action. We computed the reliability of each voxel’s action-specific activity across cross-validation folds. We then clustered voxels into action sensitive and non-sensitive classes based on these reliability values. The displayed results indicate the overlap of the selected voxels across cross-validation folds, with more consistently selected voxels shown in darker orange.