Title
Towards trustworthy AI systems: security of multi-model and multi-agent AI
Abstract
AI is rapidly evolving from individual models to systems composed of multiple interacting models and agents. In this talk, I will briefly overview my research journey from computer vision and human-centred visual understanding to robustness, security, privacy, and trustworthy foundation models. I will then focus on two recent works that explore emerging security challenges at the system level. First, I will discuss wide-net-casting attacks that exploit multiple large models, highlighting how attacks can extend beyond individual models. I will then present our work on strategic attacks in monitored multi-agent discussions, where an attacker exploits the interaction and influence structure among agents while attempting to remain stealthy. Together, these works illustrate how the security problem changes as AI evolves from individual models to interconnected AI systems.
Bio
Professor Hossein Rahmani, Lancaster University
Hossein Rahmani received the Ph.D. degree from the University of Western Australia, the MSc degree from Shahid Beheshti University, and the BSc degree from Isfahan University of Technology. His research interests include computer vision and machine learning. His works have been published in premier computer vision journals and conferences, including TPAMI, CVPR, ICCV, and ECCV. He received the Best Scientific Paper Award from the International Conference on Pattern Recognition. He is an Associate Editor of IEEE Transactions on Neural Networks and Learning Systems, Pattern recognition, ACM Computing Surveys , and Area Chair of CVPR 2025, CVPR 2024, ICLR 2025, ECCV 2024, IIJCAI 2023 and IJCAI 2024.
- Google Scholar: https://scholar.google.com/citations?user=zFyT_gwAAAAJ&hl=en
- Personal Webpage: https://sites.google.com/view/rahmaniatlu
Practical
The Tuesday Seminar Series covers a range of topics relevant to Integreat’s research focus. Presenters from within and beyond the Integreat community give 40-minute presentations, followed by group discussions.
Speaker is presenting in person in Oslo.
Seminars are open.
For those unable to attend in person: https://uio.zoom.us/j/65632027408
Points of contact:
- Organiser and moderator: Pooya Zakeri, University of Oslo
- Integreat Administrative Leader Maria Dikova, University of Oslo