Masked Modeling for Self-supervised Representation Learning on Vision and Beyond