Prof. Yang Xiang from Swinburne University of Technology, IEEE Fellow
Group photo of Prof. Yang Xiang and Prof. Haibo Hu
The RCPST Distinguished Lecture Series was successfully held on 21 November 2024, featuring an insightful presentation by IEEE Fellow Prof. Yang Xiang, Director of the Digital Capability Research Platform at Swinburne University of Technology, Australia. The event was hosted by Professor Haibo Hu, Director of RCPST and Associate Head of the Department of Electrical and Electronic Engineering at PolyU. His opening remarks set the stage for a thought-provoking session on the critical topic of AI security.
Prof. Xiang delivered a distinguished lecture titled "Securing AI Systems: From Development to Deployment." He began by emphasizing the rapid advancements in artificial intelligence (AI) systems, which have become integral to numerous applications. This integration underscores the urgent need for innovative security measures to protect these systems throughout their lifecycle. Prof. Xiang highlighted that, despite their high accuracy, AI models are inherently vulnerable to a range of sophisticated attacks, posing significant risks to their security and trustworthiness. He stressed the importance of addressing the comprehensive security challenges faced by AI systems from their development phase through to deployment.
Following this introduction, Prof. Xiang delved into the advanced defensive methods developed by his research team. These methods are designed to detect and neutralize sophisticated backdoor attacks that threaten the integrity of AI systems during their development. He also presented findings on the vulnerabilities of video classification systems, introducing a novel adversarial attack that underscores the threats to AI model availability during deployment. To conclude the lecture, Prof. Xiang offered valuable advice on the accountability of AI code generators from both development and deployment perspectives. His insights provided a comprehensive understanding of how to ensure accountability throughout the AI system lifecycle, emphasizing the need for robust security frameworks and ethical considerations in AI development.
The lecture was followed by an engaging Q&A session, where attendees had the opportunity to discuss these pressing issues further with Prof. Xiang. On behalf of the RCPST, Prof. Hu then presented Prof. Xiang with a souvenir to express gratitude for his valuable contribution to the lecture series and his insights into the critical field of AI security.
The event was a resounding success, leaving participants with a deeper understanding of the complexities involved in securing AI systems and the innovative solutions being developed to address these challenges.
