Recently, a research team from the School of Materials Science and Engineering at Xi'an Jiaotong University published a paper titled "Materials Informatics: From Germination to Autonomous Discovery in the Age of Artificial Intelligence" in the prestigious journal Advanced Materials. This comprehensive study offers a systematic overview of the evolution of materials informatics, tracing its trajectory from the introduction of the aperiodic crystal concept in 1944 to the present day. Special emphasis is placed on the role of active learning methodologies—such as Bayesian optimization and reinforcement learning—in materials design. The paper underscores the unique advantages of reinforcement learning in navigating high-dimensional search spaces, a critical challenge in modern materials research.

Furthermore, the study evaluates the transformative potential of Transformer-based large language models (LLMs) in materials science. By comparing domain-specific models like SteelBERT with general-purpose counterparts such as DeepSeek and Gemini, the research reveals significant advancements in few-shot learning and reasoning capabilities within LLMs. These breakthroughs hold promise for accelerating materials discovery and optimization processes.

Looking ahead, the paper envisions a future where "AI scientists" and "virtual laboratories" play pivotal roles. It highlights the importance of integrating active learning, uncertainty quantification, retrieval-augmented generation (RAG), and AI agents to build autonomous driving laboratories—a vision that could revolutionize the field. This groundbreaking work was conducted by teams from the National Key Laboratory for Strength of Metallic Materials and the National Key Laboratory for Porous Metal Materials. Under the expert guidance of Academician Sun Jun and Professor Ding Xiangdong, the research involved contributions from doctoral student Liu Yujie, with Associate Professor Gao Zhibin serving as the corresponding author.