Core Insights - The research team at Kunming University of Science and Technology has made significant breakthroughs in the intersection of "Artificial Intelligence + Materials" by proposing a "Continuous Transfer" machine learning framework, addressing the technical bottleneck of multi-performance prediction of materials with small datasets [1][2] - The framework allows for the efficient development of new functional materials, demonstrating the universality of transfer learning in optimizing multiple material properties [2] Group 1: Research Achievements - The team successfully constructed a "Continuous Transfer" learning strategy that first trains a high-precision base model using extensive formation energy data, followed by sequential predictions of key material properties such as stability, bandgap, and bulk modulus [1] - In a shear modulus prediction task with only 51 data points, the team utilized a bulk modulus model as a "stepping stone" for secondary transfer, significantly enhancing prediction reliability in small datasets [1] Group 2: Material Discovery - Using the framework, the research team rapidly screened over 18,000 candidate materials, identifying 54 inorganic double perovskite coating materials with high stability and excellent ductility [2] - Among these, cesium copper hexafluoroiridate exhibited outstanding performance, with a bandgap suitable for photovoltaic applications and a high ductility indicated by the ratio of shear modulus to bulk modulus [2] Group 3: Implications for the Industry - This research not only provides a candidate material library for fields such as perovskite solar cells and photocatalysis but also offers a scalable computational tool to address the challenges of data scarcity in material development [2] - The advancements in material informatics signify a crucial step in solving the "few data, many tasks" dilemma in material research, providing an efficient computational paradigm for multi-performance optimization [2]