【摘要】 依靠传统试验-试错方法已无法满足当前对于高性能材料的快速研发及定向设计需要，机器学习作为人工智能的一个重要分支，通过在输入数据与目标性能之间建立映射关系，为材料的设计提供有效指导，从而缩短材料的开发周期。结合前沿研究成果，总结出材料科学领域中机器学习的构建方法；针对数据少、数据集规模小、样本不平衡等问题，材料科学领域中数据扩充的方法主要为从出版物中提取数据、构建数据库、进行数据源层面的高通量试验和计算；小数据机器学习策略主要为主动学习、迁移学习和叶贝斯优化。机器学习在材料科学领域中机遇与挑战并存。更多还原

【Abstract】 Traditional trial-and-error experimental approaches are no longer sufficient to meet the current demands for rapid development and targeted design of high-performance materials. Machine learning,as an important branch of artificial intelligence,provides effective guidance for material design by establishing a mapping relationship between input data and target performance,thereby shortening the cycle of material development. Based on recent advances,the construction methodologies of machine learning in materials science are summarized. To address challenges such as limited data availability,small dataset size and sample imbalance,data expansion strategies in materials science mainly include extracting datum from publications,constructing specialized databases and performing high-throughput experiments and computations at the data-source level. Small-data machine learning strategies primarily involve active learning,transfer learning and Bayesian optimization. The application of machine learning in materials science presents both significant opportunities and notable challenges.更多还原