Abstract: Traditional anchor graph clustering (AGC) methods usually perform suboptimally when dealing with subspace similarity caused by spectral mixing and typically lack physical interpretability ...