The Institute of High Energy Physics, the Institute of Modern Physics, and the Hefei Institutes of Physical Science of the Chinese Academy of Sciences, along with multiple other research institutions and universities, have jointly published a review article systematically examining the technological application pathways of artificial intelligence in particle and nuclear physics. The article focuses on areas such as autonomous accelerator control, intelligent detector upgrades, real-time processing of massive data, experimental event reconstruction, theoretical model computation, and intelligent operation of large-scale scientific facilities, with a key emphasis on how AI models can be integrated into experimental systems and scientific workflows.

In accelerator systems, AI is primarily used for beam state identification, parameter optimization, anomaly detection in operations, and autonomous control. Accelerator operation involves numerous devices such as magnets, power supplies, radio frequency systems, vacuum systems, beam diagnostics, and control systems. Beam position, energy spread, emittance, brightness, and stability are influenced by the coupling of multiple parameters. Machine learning models can establish a mapping relationship between beam states and control parameters based on historical operational data and real-time diagnostic signals, aiding in the adjustment of magnetic fields, voltages, phases, and injection conditions, thereby reducing the need for manual trial-and-error tuning.

Intelligent detector upgrades focus on signal identification, noise suppression, event reconstruction, and online filtering. Detectors in particle and nuclear physics experiments often need to operate under conditions of high counting rates, high backgrounds, and complex signal overlaps. Traditional algorithms rely heavily on manually set features and thresholds. Deep learning models can process image, waveform, trajectory, and time-series data output by detectors to identify particle tracks, energy deposits, vertex positions, and candidate signals for rare events.

Real-time processing of massive data is a key technological scenario emphasized in the article. High-energy physics experiments generate high-frequency data streams, and raw data cannot all be stored long-term. Rapid decisions must be made during the triggering, compression, filtering, and reconstruction stages. AI models can be embedded in online trigger systems to quickly classify detector signals, preemptively discard low-value background events, and retain data more likely to contain target physical processes. For backend analysis, AI can also be used for event classification, parameter fitting, error estimation, and background modeling, improving the processing efficiency of complex data samples.

The article also discusses the role of AI in theoretical calculations and simulations. Research in particle and nuclear physics requires extensive Monte Carlo simulations, reaction cross-section calculations, nuclear structure models, transport models, and many-body system computations. Machine learning methods can be used to accelerate simulation generation, replace some high-cost numerical calculations, build surrogate models, and search for optimal model parameters in high-dimensional parameter spaces. For the experimental design phase, AI can also participate in optimizing detector geometry, screening experimental conditions, and adjusting data acquisition strategies.

In systems related to synchrotron radiation light sources, neutron sources, nuclear science facilities, and nuclear power, AI applications are more oriented towards facility control, state diagnostics, and predictive maintenance. Models can read equipment operational status, sensor data, image data, and experimental process data to identify equipment drift, abnormal fluctuations, fault precursors, and changes in operational efficiency. For large-scale scientific facilities requiring long-term continuous operation, such technologies can be integrated into control rooms, equipment monitoring systems, and experimental scheduling systems to participate in online feedback and operational parameter adjustments.

This review breaks down AI technology into multiple specific stages of particle and nuclear physics experiments: the front-end acquisition side handles detection signal identification and trigger filtering; the intermediate processing side handles event reconstruction, data compression, and feature extraction; the backend analysis side handles classification, fitting, simulation acceleration, and model inference; and the facility operation side handles beam tuning, equipment diagnostics, and experimental workflow optimization. The entire technological roadmap revolves around the synergy of data, models, control systems, and large-scale scientific facilities.