Can machine learning identify the next high-temperature superconductor? Examining extrapolation performance for materials discovery,Molecular Systems Design & Engineering

当前位置： X-MOL 学术 › Mol. Syst. Des. Eng. › 论文详情

Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)

Can machine learning identify the next high-temperature superconductor? Examining extrapolation performance for materials discovery
Molecular Systems Design & Engineering ( IF 3.2 ) Pub Date : 2018-08-17 00:00:00 , DOI: 10.1039/c8me00012c
Bryce Meredig _{1,

2} , Erin Antono _{1,

2} , Carena Church _{1,

2} , Maxwell Hutchinson _{1,

2} , Julia Ling _{1,

2} , Sean Paradiso _{1,

2} , Ben Blaiszik _{2,

2,

3,

4} , Ian Foster _{2,

2,

3,

4} , Brenna Gibbons _{2,

5} , Jason Hattrick-Simpers _{2,

6} , Apurva Mehta _{2,

7} , Logan Ward _{2,

2,

3,

4}

Affiliation

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.

中文翻译：

机器学习能否识别出下一个高温超导体？检查外推性能以发现材料

传统的机器学习（ML）指标高估了材料发现的模型性能。我们引入（1）留一集群交叉验证（LOCO CV）和（2）一个简单的最近邻基准，以表明发现应用程序中的模型性能在很大程度上取决于问题，数据采样和外推法。我们的结果表明，ML引导的迭代实验在发现突破性材料（例如具有ML的高T _c超导体）方面可能胜过标准的高通量筛选。

更新日期：2018-08-17

点击分享查看原文

点击收藏

公开下载

阅读更多本刊最新论文