Active Learning-Assisted High-Throughput Construction of Multi-Component Non-Equilibrium Phase Maps

Materials featuring non-equilibrium phases, such as metallic glasses, have attracted intense interest in recent years owing to their outstanding mechanical, physical, and chemical properties. A non-equilibrium phase map is highly desirable in accelerating the development of viable synthesis strategies for these materials. However, the rapid construction of such a map remains challenging due to the necessity of a comprehensive survey of a vast search space. While high-throughput combinatorial experimental techniques offer an accelerated route, their efficiency can be drastically improved by coupling them with active learning. In this work, we propose an active learning framework that integrates high-throughput experimental methods with cellular automata (CA) algorithms for the rapid generation of non-equilibrium phase maps. The framework is demonstrated on high-throughput experimental datasets from five multi-component material systems. It constructs accurate phase maps using only a fraction of the data points required by exhaustive grid traversal, achieving 100% accuracy, and exhibits excellent generalizability and robustness.