No Hard Negatives Required: Concept Centric Learning Leads to Compositionality without Degrading Zero-shot Capabilities of Contrastive Models

No Hard Negatives Required: Concept Centric Learning Leads to Compositionality without Degrading Zero-shot Capabilities of Contrastive Models explores A new method for vision-language models that improves compositional understanding without sacrificing zero-shot performance, with code available.. Commercial viability score: 7/10 in Vision-Language Models.