Attacks And Defenses In Privacy-Preserving Representation Learning