Machine learning is transforming the control of particle accelerators, enabling "autonomous driving" for these complex systems. Researchers from the Institute of Modern Physics and Xiamen University have developed innovative solutions to improve accelerator operation, reducing manual intervention and enhancing efficiency. By integrating reinforcement learning and virtual accelerators, they achieved seamless transfer to real-world applications, marking a significant breakthrough. This research paves the way for more efficient, intelligent control technologies in particle accelerators and sets a milestone in AI's application to advanced scientific tools.

In a recent paper published in National Science Review, a jointed team of scholars from China and UK has made significant progress in understanding the genesis of volcanic rocks in subduction zones. By analyzing the magnesium and boron isotopic compositions of volcanic rocks and forearc serpentinites from the South Sandwich Island arc, a globally representative island arc, the team proposed a novel mechanism for island arc magma formation. Their findings suggest that island arc volcanic rocks can originate from the partial melting of serpentinite-dominated mélanges. This breakthrough significantly advances the understanding of the processes and mechanisms of material recycling in subduction zones.