2025-06-24: GPU Hours Granted on Hypothesis Generation by Oak Ridge Leadership Computing Facility

In collaboration with Oak Ridge National Laboratory, the LAMP-SYS Lab was granted 20,000 node hours on the Frontier supercomputer cluster and 2500 node hours on the Andes supercomputer cluster. This joint proposal is a collaboration with Dr. Tirthankar Ghosal, a scientist at the Advanced AI Methods at Scale (AAIMS) group in the National Center for Computational Sciences, Oak Ridge National Laboratory, my student Dominik Soós, and me. 

The goal of this proposal is to investigate the feasibility of generating hypotheses through interactions between expert LLMs and then ranking hypothesis candidates by Z-scores, a novelty metric developed by Dr. Uzzi. 

The project will advance hypothesis generation by improving the hypothesis novelty in the candidate generation phase and the candidate selection phase. The introduction of Z-scores provides a more scalable way to automatically evaluate novelty. The multi-agent LLMs will have the potential to mitigate the consistent mistakes and biases made by a single LLM. Existing research mostly relies on data from one narrow domain (e.g., ACL). The introduction of multiple expert LLMs will overcome the limitation by generating cross-domain hypotheses. We plan to use this exploratory effort to study the scaling efficiency of our methods for larger science datasets that cover most STEM domains (e.g., with scale - Web of Science or CiteSeerX, etc.) and are crucial to leverage cross-disciplinary knowledge for scientific discovery. 

Dr. Jian Wu, the PI of this project, is an associate professor of Computer Science at Old Dominion University, Norfolk, VA. His research interests include natural language processing, scholarly big data, information retrieval, digital libraries, and the science of science.

Dr. Tirthankar Ghosal is a Staff Scientist at Oak Ridge National Laboratory’s National Center for Computational Sciences (NCCS) and will serve as Co-PI on this project. His research expertise spans AI for Science and Operations, Natural Language Processing, Large Language Models, and Machine Learning.

Dominik Soós is a PhD student of Computer Science at Old Dominion University, where he also received his B.S. and M.S. degrees in Computer Science. His research interests include natural language processing, machine learning, and parallel computing. 

-- Jian Wu