QUEST: Quadruple Multimodal Contrastive Learning with Constraints and Self-Penalization | Read Paper on Bytez