Researchers from the University of California, San Diego, have developed KeplerAgent, an innovative framework employing large language models (LLMs) to automate the discovery of equations governing physical phenomena. This system mimics scientists by leveraging a multi-step reasoning process, improving symbolic regression methods like PySINDy and PySR. Unlike traditional methods, KeplerAgent identifies structural properties, such as symmetries, guiding the search for equations more effectively. Demonstrating substantial improvements, it achieved an 89.7% success rate in recovering equations and a remarkable R-squared value of 0.992 across benchmarks. KeplerAgent also displayed resilience against noise, maintaining 82.1% accuracy with data perturbation, setting a new standard in equation discovery. While overcoming the limitations of existing methods, KeplerAgent highlights the significant potential of integrating AI with scientific inquiry, paving the way for advancements in various fields, including physics and climate science. Future developments aim to enhance its reasoning capabilities, further facilitating intelligent data analysis and discovery.
Source link
Share
Read more