Fig. 2. Closed-loop optimization for the discovery of quaternary metallic SINPs.

(A to C) Iterations 1 to 3, respectively, of the closed feedback loop process. Starting from available input data, the top compositions suggested by the machine learning algorithm are tabulated and then synthesized by SPBCL. The resulting experimental compositions, as determined by STEM-EDS, are then fed back into the algorithm for the next set of predictions. Bar graphs show elemental compositions determined by EDS averaged across multiple NPs, with error bars showing the SD and black bars representing the suggested targets. Below are annular bright-field (ABF) images and EDS maps of a representative particle for each composition. Scale bars, 20 nm. Note that particle C2 contains a small amount of Cu contamination. at %, atomic %.