Berkeley Lab’s Science IT team won the Best Industry Track Paper award at the 4th International Conference on AI/ML Systems, which took place on October 8-11, 2024 in Baton Rouge, LA. The paper, titled Aggregated Knowledge Model: Enhancing Domain-Specific QA with Fine-Tuned and Retrieval-Augmented Generation Models, was inspired by the work of Science IT consultant Fengchen Liu during an AI/ML program at Stanford. The research team, led by Liu, also included Jordan Jung, Wei Feinstein, Jeff D’Ambrogia, and Gary Jung.
The paper introduces an innovative approach to using AI models for answering complex scientific questions. Usually, different AI models, like ChatGPT or Gemini, give various answers to the same question. Liu proposed an aggregated knowledge model (AKM) that takes all these different answers, evaluates them, and then presents the most accurate and reliable one to the user. This method helps scientists by saving time by offering a suggestion for the best possible answer.
Gary Jung, ScienceIT Department Head at the Berkeley Lab explained, “Our goal was to create a system that reduces confusion for researchers by giving them an option for a single, well-supported answer. By using multiple AI models and aggregating their responses, we’re able to provide scientists with more reliable information, which can make a big difference in their work.”
The team’s work is particularly important for domain-specific research, where accuracy is critical, and even small errors can lead to wasted time or misleading results. The model they developed helps to streamline the question-and-answer process in scientific research, ensuring users get the most precise and useful information possible.
AI’s Increasing Importance in Science Research
This achievement highlights the increasing role that AI and machine learning play in advancing scientific research. The Science IT team’s success in creating this model is a significant step forward in improving how scientists interact with AI tools. Their award-winning work is a clear example of how AI can enhance research by providing faster, more accurate information to those who need it most.