Dr. Zhang Zhuosheng, a graduate from the Department of Computer Science and Engineering, Shanghai Jiao Tong University, has been recognized with the 2023 Outstanding Ph.D. Dissertation Award by the Chinese Information Processing Society of China (CIPS) for his dissertation titled 'Exploring More Reliable Pre-trained Language Models'.
The Chinese Information Processing Society of China is a national-level organization. It aims to promote the interdisciplinary integration and development of computer science, linguistics, artificial intelligence and other relevant disciplines. To advance technological progress in information processing, encourage innovative research, and foster the growth of young talents in China, the selection process for the 2023 Outstanding Ph.D. Dissertation Award commenced in June. After material submission, format review, initial expert evaluation, public disclosure of preliminary results, and final expert review, eight dissertations were chosen as the "Outstanding Ph.D. Dissertation" and four were nominated.
Dr. Zhang Zhuosheng Supervisor：Professor Zhao Hai
Dr. Zhang's dissertation focuses on the reliability of pre-trained language models, delving into two aspects: the reliability of knowledge basis and the reliability of reasoning processes. The goal is to achieve high-performance, reliable, and interpretable pre-trained language models. Addressing the demand for reliable knowledge basis in knowledge acquisition by large models, the thesis proposed a syntactic parsing enhanced model and a vision retrieval enhanced model. Starting from the perspectives of internal
linguistic knowledge parsing and external world knowledge retrieval, these two models can provide knowledge basis of linguistics the the world for language modeling so that to enhance the ability to represent language structure and contextual knowledge.To meet the demand for knowledge verification and cognitive explanation during knowledge application by large models, the thesis proposed retrospective answerability verification technique and automatic chain-of-thought prompting technique. These enhance the ability to reject inappropriate requests and engage in step-by-step reasoning, effectively verifying and explaining the reasoning process. The research findings have been published in top journals and conferences, including IEEE TPAMI, ACL, ICLR, and AAAI.
Framework of the dissertation
Dr. Zhang Zhuosheng's academic contributions have garnered over 3200 citations on Google Scholar. During his doctoral studies, he received honors such as the 2023 World Artificial Intelligence Conference Yun Fan Award Tomorrow's Star, the 7th Shanghai Jiao Tong University Academic Star, Baidu Scholarship, and recognition as one of the Global Top 100 Chinese AI Scholars. Dr. Zhang has joined the School of Cyber Science and Engineering of Shanghai Jiao Tong University as a tenure-track assistant professor, where he will continue to conduct in-depth research on pre-trained models and security.