New data analysis methods at the Beijing Spectrometer Ⅲ

Since the discovery of the J/ψ particle in 1974, the past half-century has witnessed remarkable advancements in accelerator and detector technology, leading to a many orders-of-magnitude increase in both the volume and complexity of data collected in high-energy physics experiments. Today, experiments at the Beijing Spectrometer Ⅲ (BESⅢ) have accumulated more than 10 billion J/ψ events, marking an 8 orders-of-magnitude improvement in statistics compared with the original J/ψ discovery by Samuel C. C. Ting. Concurrently, the methodologies for analyzing experimental data have undergone significant evolution, driven by breakthroughs in computational technologies and algorithms. From traditional statistical methods to boosted decision trees and deep learning, researchers have continuously explored more efficient ways to rapidly and accurately extract physical information from the large datasets. Within the BES Ⅲ experiment, advanced multivariate analysis and machine learning models have been employed for detector simulation, charged track reconstruction, particle identification, and event selection, significantly enhancing experimental sensitivity and efficiency. In recent years, the emergence of large language models with their powerful text and code generation capabilities has opened new possibilities for automated physics analysis. Inspired by this, researchers have developed the AI agent“Dr. SAI”to accelerate the extraction of physical results in high-energy physics experiments. This approach is expected to make a substantial impact on high-energy physics research and may potentially usher in a transformative paradigm shift in scientific discovery.