The National Science Foundation (NSF) has awarded a $299,756 Project Grant under the Mathematical and Physical Sciences (CFDA 47.049) program to the University of California, Berkeley. The funding will support a theoretical study on improving the efficiency and generalization performance of fine-tuning and transfer learning techniques for large language models like ChatGPT. The research aims to develop methods for optimizing hyperparameters during fine-tuning and to analyze transfer learning in over-parameterized linear regression models. The project will provide financial support and mentorship for graduate students and is expected to run from August 1, 2024 through July 31, 2027. The findings have the potential to significantly impact the practical use of large language models across scientific research, education, medicine, government, and business applications.
Generated 1/28/25, 10:15 AM