AIMATDESIGN: knowledge-augmented reinforcement learning for inverse materials design under data scarcity

With the growing demand for novel materials, machine learning-driven inverse design methods face significant challenges in reconciling the high-dimensional materials composition space with limited experimental data. Existing approaches suffer from two major limitations: (I) machine learning models often lack reliability in high-dimensional spaces, leading to prediction biases during the design process; (II) these models fail to effectively incorporate domain expert knowledge, limiting their capacity to support knowledge-guided inverse design. To address these challenges, we present AIMATDESIGN, a reinforcement learning framework that augments experimental data via difference-based sampling to build a trusted experience pool. To enhance model reliability, an automated refinement strategy guided by large language models corrects prediction inconsistencies, ensuring better alignment between reward signals and value functions. Additionally, a knowledge-based reward further leverages expert rules to improve stability and efficiency. Experiments show that AIMATDESIGN outperforms conventional machine learning and reinforcement learning baselines in discovery efficiency, convergence speed, and success rates. From numerous candidates, experimental validation of Zr-based alloys yielded a bulk metallic glass with 1.7 GPa yield strength and 10.2% elongation, closely matching predictions. This framework demonstrates reliable performance in capturing property trends and provides a general pathway toward closed-loop, data-efficient materials discovery.