Tighter generalization bounds for matrix completion based on non-negative matrix factorization with norm constraints (情報論的学習理論と機械学習)