Learning Robust and Privacy-Preserving Representations via Information Theory