Roadmap on machine learning glassy dynamics

Unravelling the connections between microscopic structure, emergent physical properties and slow dynamics has long been a challenge when studying the glass transition. The absence of clear visible structural order in amorphous configurations complicates the identification of the key physical mechanisms underpinning slow dynamics. The difficulty in sampling equilibrated configurations at low temperatures hampers thorough numerical and theoretical investigations. We explore the potential of machine learning (ML) techniques to face these challenges, building on the algorithms that have revolutionized computer vision and image recognition. We present both successful ML applications and open problems for the future, such as transferability and interpretability of ML approaches. To foster a collaborative community effort, we also highlight the ‘GlassBench’ dataset, which provides simulation data and benchmarks for both 2D and 3D glass formers. We compare the performance of emerging ML methodologies, in line with benchmarking practices in image and text recognition. Our goal is to provide guidelines for the development of ML techniques in systems displaying slow dynamics and inspire new directions to improve our theoretical understanding of glassy liquids. Slow heterogeneous dynamics and the absence of visible structural order make it difficult to numerically and theoretically investigate glass-forming materials. This Technical Review outlines the role that machine learning tools can have and identifies key challenges, possible approaches and appropriate benchmarks. Systematic characterization of amorphous glassy structures can be addressed by unsupervised learning, which requires an adequate choice of structural descriptors. Finding structure–dynamics relationships in glassy liquids is a task that has many analogies with image recognition and can be tackled using supervised learning with various neural network architectures already successful in image recognition. Major challenges and potential breakthroughs await in transferring trained models to extremely low temperatures, using them to create ultrastable glasses and design new phenomenological glass models. Future directions also encompass generative modelling of low-temperature equilibrium configurations and development of self-supervised and reinforcement learning approaches. Publicly available datasets and unified benchmarks that are fundamental to stimulate further development of ML techniques in condensed matter physics are provided. Systematic characterization of amorphous glassy structures can be addressed by unsupervised learning, which requires an adequate choice of structural descriptors. Finding structure–dynamics relationships in glassy liquids is a task that has many analogies with image recognition and can be tackled using supervised learning with various neural network architectures already successful in image recognition. Major challenges and potential breakthroughs await in transferring trained models to extremely low temperatures, using them to create ultrastable glasses and design new phenomenological glass models. Future directions also encompass generative modelling of low-temperature equilibrium configurations and development of self-supervised and reinforcement learning approaches. Publicly available datasets and unified benchmarks that are fundamental to stimulate further development of ML techniques in condensed matter physics are provided.