Title: Beyond Facts - Understanding and Inducing Rule-based Knowledge in LLMs

Abstract: Large language models (LLM) have been proven to be able to learn knowledge from massive data. Most research currently discusses the relationship between implicit knowledge in LLMs and symbolic factual knowledge in Knowledge Graphs. Besides facts, human knowledge contains more types, such as rules. How does a LLM understand a rule and promote reasoning ability? Whether a LLM induce new rules from the given data? This talk will introduce our latest research work on these questions.

Bio: Kang Liu is a full professor at Institute of Automation, Chinese Academy of Sciences. He is also a youth scientist of Beijing Academy of Artificial Intelligence and a professor of University of Chinese Academy of Sciences. His research interests include Knowledge Graphs, Information Extraction, Question Answering and Large Language Models. He has published over 80 research papers in AI conferences and journals, like ACL, EMNLP, NAACL, COLING, et al. His work has over 20,000 citations on Google Scholar. He received the Best Paper Award at COLING-2014, Best Poster&Demo Award at ISWC-2023, and the Google Focused Research Award in 2015 and 2016.