Highlights of recent machine learning applications in nuclear physics

Machine learning has emerged as a powerful tool for data analysis, prediction, modeling, and simulation across nuclear physics. While early applications of deep learning focused on narrow tasks where one had access to large labeled datasets, such as particle identification and event reconstruction, recent advances in general-purpose, flexible models are beginning to transform the field. In this talk, I will highlight recent work leveraging advanced machine learning architectures tailored for nuclear physics data, including symmetry-preserving graph neural networks, point cloud models, sparse networks, and Kolmogorov-Arnold Networks. I will also discuss efforts in uncertainty quantification and stochastic modeling, which are essential for evaluating model reliability and robustness in scientific contexts. Finally, I will highlight emerging "foundation models" for physics data -- large, pre-trained models that can be adapted to diverse downstream tasks such as classification, regression, reconstruction, and simulation. These developments suggest a path toward more scalable, reusable tools that can support a broader community of nuclear physicists.