Jul 2026
How Do VLMs Fail? Vision-Operation Misalignment in Compositional VQA
An Operation-centric mechanistic framework is introduced that decomposes VLM failures by both the reasoning operation where they originate and the internal computational pathway through which they propagate and provides a principled basis for targeted diagnosis of VLM failures in multimedia reasoning.
Navya Gupta, Bingjie Xu, Avinash Anand et al.
· arXiv.org · 0 citations