The Data Science discipline in the School of EECS is hosting the following guest research seminar
Towards Trustworthy AI
Speaker: Dr Qiongkai Xu
Host: Dr Yadan Luo
Abstract: Artificial Intelligence (AI) has showcased its superior capability in various domains, particularly with large-scale pre-trained models. Despite these impressive achievements, many AI applications still face limitations and restrictions due to ethical, technical, and social concerns that need to be addressed before we can fully trust these systems. In this seminar, Dr. Qiongkai Xu will present his research towards trustworthy AI systems, focusing on three key areas. Firstly, he will introduce a new evaluation theorem that is able to certify 'superhuman' performance of classifiers. Secondly, he will present his doctoral research on preventing privacy leakage in conversations. Lastly, he will discuss his work on attack and defence strategies, with a specific focus on imitation attack and watermarking techniques for natural language processing models.
About Data Science Seminar
This seminar series is hosted by EECS Data Science.