Supplementary Material: Distribution Aligning Refinery of Pseudo-label for Imbalanced Semi-supervised Learning A Proof of Theorem 1, r 2 R n 0, c 2 R m 0