What Machine Learning Methods is Physics invested in?

The intersection of machine learning (ML) and physics is rapidly expanding, yet a comprehensive overview of which ML techniques are being adopted by physicists – and within which sub-disciplines – remains lacking. This work addresses this gap by presenting a large-scale bibliometric analysis of all accepted papers from the Machine Learning for Physical Sciences (ML4PS) workshop at NeurIPS over the past years. We leverage Large Language Models (LLMs) to automatically classify each publication according to both its primary physics discipline and the employed ML task or methodology. This enables us to identify dominant trends in ML application across physics, and to track the evolution of these trends over time. Our results provide a quantitative snapshot of the current ML landscape within the physics community, highlighting prevalent approaches and emerging areas of research.