Associate Professor Yun Huang, Assistant Professor Jiaqi Ma, and Assistant Professor Haohan Wang have received computing resources from the National Artificial Intelligence Research Resource (NAIRR), a two-year pilot program led by the National Science Foundation in partnership with other federal agencies and nongovernmental partners. The goal of the pilot is to support AI-related research with particular emphasis on societal challenges. Last month, awardees presented their research at the NAIRR Pilot Annual Meeting.
Huang received $15,000 in computational resources for her project, "Harnessing Large Language Models for Ethical Research and Innovation." AI systems can benefit students by boosting their academic performance, stimulating their curiosity, and making learning more efficient. This is especially important for learners with diverse needs. According to Huang, the project focuses on facilitating the discovery of insightful questions given learning materials to guide and educate learners in scaffolding and creating learning ideas. The goal is to introduce new systems and methods using natural language generation models to augment a learner's discovery process. "The computing support allows us to launch our systems and provide scalable services. It is exciting that over 330 people from about 20 organizations have been using our tools," Huang said.
Ma received 4,000 node hours for TACC Vista (NVIDIA GH100 Grace Hopper Superchip) for his project, "Principled Quantification of Training Data Influence on Generative AI Models." This research aims to quantify the influence of individual training data points on generative AI models, which will help address AI safety issues concerning privacy, copyright, and hallucination. According to Ma, current methods rely heavily on model retraining, leading to ambiguity in evaluation metrics and high computational costs—especially for large models. His team will conduct research to understand and effectively simulate the randomness inherent in training dynamics, thereby developing more efficient techniques to accurately measure training data influence.
Wang received $90,000 in computing resources for his project, "Toward Virtual Bioinformaticians: Empowering Scientific Discovery in Genomics with LLM-Based Agents." Bioinformaticians handle the management, processing, and analysis of large sets of genomic and molecular data. These specialists play a crucial role in biomedical research, helping scientists examine gene activity and understand biological processes. According to Wang, studies show that bioinformaticians spend nearly half their time on routine data processing, which is not only inefficient but also costly for the industry. He proposes the Rational AI-Operated Bioinformatics Assistant Team (ROBAT)—a system of AI-powered assistants, or "AI Agents," designed to automate routine bioinformatics tasks, from raw data processing to producing visualized results. This will free up bioinformaticians to focus on more complex and innovative research, accelerating discoveries in genomics and beyond, he said.