Can machine learning identify the next high-temperature superconductor? Examining extrapolation performance for materials discovery. Issue 5 (17th August 2018)

• Record Type: Journal Article
• Authors: Meredig, Bryce; Antono, Erin; Church, Carena; Hutchinson, Maxwell; Ling, Julia; Paradiso, Sean; Blaiszik, Ben; Foster, Ian; Gibbons, Brenna; Hattrick-Simpers, Jason; Mehta, Apurva; Ward, Logan
• Is Part Of: Molecular Systems Design and Engineering. Volume 3:Issue 5(2018)
• Page Start: 819
• Page End: 825
• Publication Date: 2018-08-17
• Subjects: Chemistry -- Molecular aspects -- Periodicals; Chemical engineering -- Molecular aspects -- Periodicals; Nanotechnology -- Periodicals; 620.5
• DOI: 10.1039/c8me00012c
• Languages: English
• ISSNs: 2058-9689

Abstract

Abstract : Traditional machine learning (ML) metrics overestimate model performance for materials discovery. Abstract : Traditional machine learning (ML) metrics overestimate model performance for materials discovery. We introduce (1) leave-one-cluster-out cross-validation (LOCO CV) and (2) a simple nearest-neighbor benchmark to show that model performance in discovery applications strongly depends on the problem, data sampling, and extrapolation. Our results suggest that ML-guided iterative experimentation may outperform standard high-throughput screening for discovering breakthrough materials like high- T c superconductors with ML.