Are vision language models robust to uncertain inputs?

Xi Wang, Eric Nalisnick

Robustness against uncertain and ambiguous inputs is a critical challenge for deep learning models. While recent advancements in large scale vision language models (VLMs, e.g. GPT4…