Comparison between brain sub-networks decomposed by auto encoder (AE) and graph auto encoder (GAE) with non-negative weight constraints and sparse encoding