related papers: recommending materials prior to their discovery

[sgbaird]

Mentioned by Tonio: https://www.nature.com/articles/s41586-019-1335-8 https://twitter.com/toniobuonassisi/status/1555616848414232578

From the article:

"Finally, we tested whether our model—if trained at various points in the past—would have correctly predicted thermoelectric materials reported later in the literature."

From @JosephMontoya-TRI: https://www.nature.com/articles/s41598-022-08413-8 mentioned in https://matsci.org/t/how-do-i-do-a-time-split-of-materials-project-entries-e-g-pre-2018-vs-post-2018/42584/8?u=sgbaird

From the article:

Campaigns performed for single-fidelity acquisition and multi-fidelity acquisition. In campaign A, we compare the results of including all or no DFT data in the seed data set and only acquire experimental data. This campaign effectively compares an “a priori” agents to a single-fidelity agents. In campaign B, the multi-fidelity agents are seeded with first 500 experimentally discovered compositions (based on ICSD58 timeline of their first publication59) and their corresponding DFT data. Both DFT and experimental data are acquired here. The single-fidelity agents are exclusively seeded with and acquire experimental data.

https://github.com/sparks-baird/xtal2png/issues/54

[JosephMontoya-TRI]

@aykol also mentioned this one - this paper is more or less explicitly focused on material discovery forecasting via the graph properties of the phase diagram, which is the first example in the literature I know of. For the curious, there's a visualization tool for this which you can see at maps.matr.io with open-source code.