Machine-learning guided design of non-precious-metal high-entropy electrocatalysts for alkaline hydrogen evolution

High-entropy alloys (HEAs) have emerged as a promising class of electrocatalysts for sustainable hydrogen production driven by renewable electricity, owing to their compositional diversity and flexible local atomic configurations. However, the vast design space and complex local environments of HEAs pose major challenges for pinpointing active sites and enabling rational catalyst design. In this work, we develop an integrated, data-driven framework to accelerate the discovery of non-precious-metal HEA catalysts for the alkaline hydrogen evolution reaction (HER). By combining high-throughput density functional theory (DFT) calculations with machine learning models enhanced by transfer learning, we screened over 25,000 surface sites to rapidly identify catalytically promising FeCoNi-based HEAs. Among them, FeCoNiCuW emerges as a leading candidate, featuring two distinct active site motifs, NiCoW and NiCuW, that are identified computationally and validated through both DFT analysis and experimental measurements. This study establishes a generalizable approach for active site discovery in compositionally complex materials and provides actionable insights for the design of efficient HER electrocatalysts. • Transfer learning-based neural networks accelerate the discovery of optimal non-precious-metal high-entropy alloys catalysts by overcoming the spatiotemporal limits of traditional calculations. • Accurate atom-level structure–performance relationships reveal NiCoW and NiCuW sites as the active sites for alkaline hydrogen evolution reaction in FeCoNiCuW high-entropy alloys. • Predicted structure–performance relationships are validated experimentally, confirming the reliability of the AI-assisted high-throughput screening strategy.