NAIRR Pilot Award for Trustworthy Vision-Language AI

NAIRR Pilot Award for Trustworthy Vision-Language AI

CARBONDALE, Ill. — November 6, 2026

School of Computing Professor Receives NAIRR Pilot Award from the U.S. National Science Foundation to Strengthen Large Vision-Language Models Against Adversarial Attacks

Dr. Khaled Mohammed Saifuddin, assistant professor in the School of Computing and director of KhaledLab, has been awarded 14,000 GPU-hours on the NCSA Delta supercomputer to support the project, “RelGuard: Closing the Relational Gap in Adversarial Defenses for Large Vision-Language Models.” The computing allocation was awarded through the U.S. National Science Foundation's National Artificial Intelligence Research Resource (NAIRR) Pilot program.

Large vision-language models, AI systems that can look at an image and answer questions about it in natural language, are rapidly being adopted in applications where mistakes carry real consequences, including medical imaging, visual question answering, and autonomous driving. Yet these models remain vulnerable to adversarial attacks: tiny, carefully crafted changes to an image, invisible to the human eye, that can cause a model to give wrong answers or bypass its safety mechanisms entirely.

Current defenses help models recover from such attacks, but they do not fully protect how a model understands a scene as a whole. RelGuard aims to close this gap by characterizing how adversarial attacks affect this deeper level of understanding and by developing a new defense framework designed to keep it intact. The team will also create a new evaluation protocol so the research community can consistently measure this kind of robustness.

The awarded computing resources will enable Dr. Saifuddin's team to run large-scale attack and defense experiments on modern vision-language models and evaluate their approach across multiple benchmarks and attack settings.

 

Media Contact: School of Computing Communications • socinfo@siu.edu