When: Thursday, May 2, 2024, 11:00 AM - 12:00 PMWhere: zoom
Speaker: Dr. Ziquan Liu who is a Lecturer in EECS.
Zoom link: https://qmul-ac-uk.zoom.us/j/81148100921Title: The Path Towards Reliable AIAbstract: The prevalent learning paradigm, known as empirical risk minimization (ERM), operates under the assumption that data is drawn from an identical distribution and primarily centres its attention on the average risk, thereby focusing predominantly on well-represented data instances. This foundational assumption, while traditionally upheld in machine learning, is increasingly recognized as a simplification that does not accurately reflect the complexities inherent in real-world data distributions, making the existing AI systems unreliable in real-world applications. It also raises concerns related to fairness, as the emphasis on average risk tends to marginalize under-represented groups, potentially exacerbating societal inequalities. In this talk, Dr. Liu will provide an overview of his research concentrated on mitigating vulnerabilities of machine learning systems arising from adversarial attacks and distribution shifts, with a focus on computer vision and its application in healthcare domains.
Bio: Dr. Ziquan Liu is now a Lecturer (Teaching & Research) at the School of Electronic Engineering and Computer Science, Queen Mary University of London. Before joining QMUL, he worked as a Postdoc Research Fellow in Machine Learning at Information, Inference and Machine Learning group, University College London from April to December of 2023. He was a PhD student at City University of Hong Kong in Video, Image, and Sound Analysis Lab under the supervision of Prof. Antoni B. Chan, and obtained his PhD in 2023. He obtained the Bachelor of Engineering degree in Information Engineering from Beihang University and a secondary Bachelor of Science degree in Mathematics from the same university, both in 2017.