<![CDATA[Machine learning is used to rapidly predict, screen, and design materials functioning in solid-state for use in a growing range of chemical spaces that are too large for traditional trial and error approaches. This article reviews how machine learning accelerates the discovery of novel solid-state materials with emphasis on three technologically important classes: topological insulators, semiconductors, and solid electrolytes. The conversation highlights data infrastructures, chemical and structural representations, graph neural networks, foundation models, high-throughput screening and generative design, and closed-loop validation. Machine learning is used in topological materials to classify the topology and to generate a potential insulator or semimetal inverse. It is used in semiconductors to predict band gaps, phase stability and optoelectronic properties in an efficient manner. It can be used to solve multi-property optimization issues such as ionic conductivity, electrochemical stability, interfacial compatibility, and synthesizability in solid electrolytes. The authors suggest that the most successful methods for discovery are based on a combination of data-driven models, density-functional theory, atomistic simulation, uncertainty quantification, and experimental feedback. Despite recent progress, there are significant challenges in data quality, transferability, interpretability, synthesis prediction, and laboratory validation. The future will rely on the ability to embed physics-driven machine learning, self-driving laboratories and foundation models in clear and reproducible materials discovery pipelines that are supported by experimental data.]]>
