Abstract:
LLMs have shown tremendous ability and potential for bridging the human-AI gap, yet, some critical shortcomings remain. One of them is hallucination, where the LLM produces incorrect answers, or invents its own facts. As a possible solution, Knowledge Graphs (KG) provide LLMs with a source of interconnected factual knowledge, which can be used to ground the responses of an LLM. What are the various ways LLMs and KGs can work together and what are the challenges in this process?
Bio: Dr. Debayan Banerjee obtained his Bachelor of Technology degree in 2009 at the National Institute of Technology, Durgapur, India. For the next 8 years, he worked in the Indian software industry, 4 of which he spent co-founding his own startup in the area of Computer Vision. In 2017, he moved to Germany, where he completed his M.Sc. degree in Computer Science at the University of Bonn in 2020, and then finished his PhD at the University of Hamburg in 2024. He is currently a Postdoctoral Researcher (Akademischer Rat) at the Leuphana University of Lüneburg in Germany. His research interests lie broadly in the area of LLMs and Natural Language Processing (NLP), and more specifically, on the topic of Knowledge Graph Question Answering (KGQA).
Webpage: https://debayan.