arxivcs.CVcs.AI2026-06-25
Bridging Vision and Language Concepts through Optimal Transport Semantic Flow
Chenyang Zhang, Anqi Dong, Guangming Zhu, Nuoye Xiong, Siyuan Wang, Lin Mei, et al.
Concept Bottleneck Models (CBMs) promise transparent reasoning by predicting through human-interpretable concepts, yet their effectiveness fundamentally depends on how well visual and textual representations are aligned or matched. Existing vision-language CBMs often rely on pre-…