We propose a self-supervised driven consistency training paradigm for histopathology image analysis that learns to leverage both task-agnostic and task-specific unlabeled data based on two strategies: ...