AI研讨会 | Reliable Multimodal Intelligence

🎤 Speaker

Hongkang ZHANG

PhD at Tsinghua University

📰 Title

Reliable Multimodal Intelligence: From Shared Information Toward Trustworthy Adaptive Intelligence

⏰ Time

15:00 -16:00, Beijing Time

📅 Date

22 Sep 2026, Tue

📍 Venue

W2-201

🤝 Online Zoom link

https://hkust-gz-edu-cn.zoom.us/j/92252495974?pwd=Jl13JU2wTF5XWO2w4tw5b0qcb2IfZb.1

• Meeting ID: 922 5249 5974

• PW: ait

Modern AI is moving beyond benchmark prediction toward systems that reason across modalities, retain useful experience, act in open environments, and adapt over time. In these settings, reliability is not determined by predictive accuracy alone. It also depends on what shared structure a system has learned, what evidence it preserves for future decisions, and whether changes to the system are warranted by the available evidence. This seminar presents a research program on reliable multimodal intelligence through three connected questions: What is genuinely shared across modalities? What evidence must remain available under finite computation when future demands are not yet known? And when is adaptation justified for a deployed model or an evolving agent? I will discuss methods for scalable nonlinear dependence learning, reusable decision-relevant evidence, and auditable adaptation under distribution shift and continual experience. Together, these directions trace a path from representation and evidence preservation to controlled system evolution. The guiding principle is to preserve the information needed to justify what the system does next. I will conclude with a broader agenda for trustworthy adaptive intelligence: systems that can continually improve and adapt in open environments while remaining reliable, auditable, and controllable.

Hongkang Zhang recently defended his Ph.D. dissertation in Data Science and Information Technology at Tsinghua University. He received his B.S. degree in Electrical Engineering from Texas A&M University and his M.S. degree in Electrical Engineering from the University of California, Santa Cruz. He was also a Visiting Student Researcher at the University of California, Berkeley. His research focuses on reliable multimodal intelligence at the intersection of information theory, multimodal learning, trustworthy AI, and adaptive intelligent systems, with particular interests in scalable dependence learning, decision-relevant evidence modeling, and trustworthy adaptation and capability evolution in open-world intelligent systems. He has published more than ten papers as first author across leading international conferences and journals, including NeurIPS and ICML. His work has also appeared in major international venues, including ICASSP, IEEE TNNLS, and IEEE JSTARS.