Assistant Professor, University of Wisconsin - Madison
1 paper at NeurIPS 2025
MxDs show that dense layers are more faithfully represented by mixtures of specialized sublayers than by sparsely activating neurons, while remaining just as interpretable.
