BREED 2.0 — ML for solid-state electrolytes '26
DFT screening is too slow, and there's almost no electrolyte data. So I used physics-based features, and combined a composition based model with a structural one to make the most of a small dataset (n=829). MAE ≈ 0.94 log σ — roughly an order of magnitude, which tells you screening, not precision.
beats the published OBELiX RF baseline by 36%