Figure 2.

a) An additive library that includes 12 metal salts (Variable 1, M) and more than 200 organic molecules (Variable 2) used to construct a set of CO₂ reduction electrocatalysts. b) The learning loop consisting of three iterative cycles of “experimental test–ML analysis–prediction and redesign” to accelerate the search for high‐performance catalysts. c) Feature importance of FE‐C₂₊ obtained by gradient boost decision tree regressor analysis in the first round of learning (right), the selected catalysts with maximal FE for different products after the second round of learning (middle), and the experimentally‐measured FE‐C₂₊ values of three selected catalysts (left). FE = Faradaic efficiency; MFF = molecular fragment featurization. Reproduced with permission.^{[ 11 ]} Copyright 2021, American Chemical Society.