NUS Computing tops challenge leaderboard at ICML Workshop

NUS Computing researchers rank first in the challenge leaderboard at the ICML Trustworthy Multi-modal Foundation Models and AI Agents Workshop, a competition featuring two main tasks designed to advance the development of trustworthy large-language models and agents.

The team includes research fellow Yangyang Guo, postdoc researcher Ziwei Xu, senior research fellow Yong Kang Wong, PhD student Xilie Xu and Professor Nie Liqiang, an alumnus of NUS Computing’s PhD program.

The leaderboard for the Track I challenge involves designing inputs that can successfully impair the accuracy of the multimodal large language model Llava-1.5. Participants alter image or text inputs in a way that significantly impacts the model’s ability to understand and respond appropriately. By designing inputs that prompt Llava-1.5 to generate incorrect or potentially harmful outputs, researchers can deepen their understanding of AI robustness and how to ensure these systems can be trusted.