学术讲座

报告题目：Trustworthy Natural Language Processing with Knowledge Guidance

讲座时间：2023年3月23日（周四）10:00-11:30

讲座地点：中国人民大学立德楼18层1826多功能报告厅

报告摘要：Large pre-trained language models (LLMs) are becoming a foundamental tool in the area of natural language processing (NLP). Almost any downstream task can be solved with only a few or without any annotated exemplars, via few-shot or zero-shot in-context learning on LLMs. We will first start with some empirical studies on ChatGPT, followed by our recent work using LLMs as knowledge source for fact checking. However, some ridiculous results trigger our contemplation "who watches the super-powerful LLMs". Then, we compare knowledge graph (KG) with LLMs and propose to combine their merits for trustworthy NLP. We will introduce two recent progress to discuss: 1) what type of knowledge needed? and 2) what role shall KG play in the era of LLMs? Finally, we will make a conclusion and share our thoughts on the future challenges.

主讲嘉宾：

曹艺馨，助理教授，研究领域为知识图谱和自然语言处理。2018年于清华大学取得博士学位，毕业后在新加坡国立大学从事博士后研究，之后在南洋理工大学担任研究型助理教授。在CCF A、B的国际权威会议和期刊发表论文30余篇，谷歌引用超过2600余次。多次担任国际顶级会议委员会委员，主持和参与多项新加坡政府项目。