Are vision language models robust to uncertain inputs?
This addresses the critical issue of model reliability for practitioners using VLMs in real-world applications, but it is incremental as it builds on existing uncertainty quantification tasks.
The study tackled the problem of robustness in vision-language models (VLMs) against uncertain inputs, finding that while larger models show improved robustness, they still hallucinate confidently on unclear inputs, but prompting them to abstain can achieve near-perfect reliability in some settings like ImageNet, though domain-specific tasks like galaxy classification remain challenging.
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. GPT4o) might suggest that increasing model and training dataset size would mitigate this issue, our empirical evaluation shows a more complicated picture. Testing models using two classic uncertainty quantification tasks, anomaly detection and classification under inherently ambiguous conditions, we find that newer and larger VLMs indeed exhibit improved robustness compared to earlier models, but still suffer from a tendency to strictly follow instructions, often causing them to hallucinate confident responses even when faced with unclear or anomalous inputs. Remarkably, for natural images such as ImageNet, this limitation can be overcome without pipeline modifications: simply prompting models to abstain from uncertain predictions enables significant reliability gains, achieving near-perfect robustness in several settings. However, for domain-specific tasks such as galaxy morphology classification, a lack of specialized knowledge prevents reliable uncertainty estimation. Finally, we propose a novel mechanism based on caption diversity to reveal a model's internal uncertainty, enabling practitioners to predict when models will successfully abstain without relying on labeled data.