Deep Learning Robustness: From A System Reliability’s Perspective
Distinguished Research Seminar Series
-
Date
30 Jul 2026
-
Organiser
Department of Industrial and Systems Engineering, PolyU
-
Time
10:30 - 12:00
-
Venue
DE402
Speaker
Prof. Xingyu Zhao
Remarks
If you have enquiries regarding E-certificate after the seminar, please contact david.kuo@polyu.edu.hk.
Summary
This talk examines deep learning robustness from a system reliability perspective, arguing that robustness should not be studied as an isolated AI model property but in relation to higher-level system claims such as safety assurance, reliability assessment, and security certification. It first reviews common robustness problem formulations, including binary robustness verification, maximum safe radius, maximum loss, and probabilistic robustness. Among these, probabilistic robustness is highlighted as especially relevant for safety-critical AI systems because it quantifies the likelihood of adversarial examples under stochastic perturbations, thereby aligning more naturally with reliability and risk reasoning in real-world operation.
The talk then distinguishes the roles of adversarial robustness and probabilistic robustness at system level. Adversarial robustness is more closely associated with security against maliciously optimized attacks, whereas probabilistic robustness better captures reliability and safety risks arising from benign operational noise, environmental uncertainty, and unsophisticated random perturbations. This distinction motivates a broader research agenda in which robustness metrics are selected according to the system-level property being claimed.
Building on this perspective, the talk presents the presenter team’s recent work on probabilistic robustness estimation, improvement and benchmarking. The talk advocates integrating probabilistic robustness into system-level safety and reliability modelling to support more meaningful assurance arguments for AI-enabled systems.
Keynote Speaker
Prof. Xingyu Zhao
Professor
School of Cyber Science and Engineering, Wuhan University, China
Dr. Xingyu Zhao, Professor and Doctoral Supervisor at the School of Cyber Science and Engineering, Wuhan University. He is a recipient of the National High-Level Young Talent Program (Overseas), an Honorary Associate Professor at the University of Warwick, an Honorary Lecturer at the University of Liverpool, a Fellow of the UK Higher Education Academy (FHEA). He received his Bachelor's and Master's degrees from Beihang University, and later pursued his PhD in Computer Science at Centre for Software Reliability, City University of London. After completing his PhD, he held positions at several Russell Group universities in the UK before joining Wuhan University full-time in 2026 as a full Professor in Safe AI. His research focuses on the reliability and safety technologies of safety-critical systems, with core interests including Trustworthy AI, Software Reliability, Formal Verification, Bayesian Inference, and Safety Analysis and Assurance. As Principal Investigator or Co-PI, he has led and contributed to securing research funding totaling £7.8 million. His leading research projects have been funded by UKRI-EPSRC, the European Union's Horizon Programme, the British Academy, among others, involving collaborations with industry partners such as Siemens, Huawei, NVIDIA, DENSO, Wayve, and NCC Group. He serves as Associate Editor for IEEE-TNNLS, Editorial Board Member for Nature-Communications AI & Computing, Area Chair for NeurIPS, and Grant Review Panel Member for UK Research and Innovation (UKRI). He has published over 70 papers in top-tier conferences and journals, including: Artificial Intelligence: NeurIPS, CVPR, ICCV, ECCV, ICML, AAAI, UAI, AAMAS, Elsevier-INFFUS, Springer-ML; Reliability & Safety: Elsevier-RESS, IEEE-TR, ISSRE, DSN, CCS, SafeComp; Software Engineering: ACM-TOSEM, ACM-TECS, IEEE-TSE, Elsevier-IST, ASE; Robotics & Autonomous Systems: Nature-CE, IEEE-TCCN, IEEE-RA-L, ICRA, IROS, ITSC, IV
You may also like
