Author Correction: Enhancing materials property prediction by leveraging computational and experimental data using deep transfer learning (Nature Communications, (2019), 10, 1, (5316), 10.1038/s41467-019-13297-w)

Dipendra Jha, Kamal Choudhary, Francesca Tavazza, Wei keng Liao, Alok Choudhary, Carelyn Campbell, Ankit Agrawal*

*Corresponding author for this work

Research output: Contribution to journalComment/debatepeer-review

Abstract

The original version of this Article contained some errors due to the presence of duplicates in one of the employed target data sets (EXP), which would have slightly overestimated model accuracy both for the baseline (training from scratch) and transfer learning. Correcting for duplicates in EXP results in small changes in the accuracy numbers, such that a lot of corrections should be done, both in the main text, tables and figures and in the Supplemenary Information file. Please find below a list of the needed corrections. The last sentence of the abstract originally reads “using an experimental data set of 1,963 observations”. The correct version states ‘1,643’ in place of ‘1,963’. The last sentence of the abstract originally reads: “the proposed approach yields a mean absolute error (MAE) of 0.06 eV/atom”. The correct version states “0.07 eV/atom” in place of “0.06 eV/atom”. The second last sentence of the last paragraph of the Introduction originally reads: “and an experimental data set containing 1963 samples from the SGTE Solid SUBstance (SSUB) database”. The correct version states “1,643 samples” in place of “1963 samples”. The last sentence of the last paragraph of the Introduction originally reads: “in particular, the proposed approach enables us to achieve an MAE of 0.06 eV/atom”. The correct version states “0.07 eV/atom” in place of “0.06 eV/atom”. The last sentence of the last paragraph of the Introduction originally reads: “against an experimental data set containing 1963 observations”. The correct version states “1,643” in place of “1963”. The last sentence of the first paragraph of the Results “Data sets” originally reads: “It is composed of 1,963 formation energies at 298.15 K”. The correct version states “1,643” in place of “1,963”. The last sentence of the second paragraph of the Results “Training from scratch” originally reads: “The impact of training data set is most evident in the case of the experimental data set, where the training data for each fold of the 10-fold cross-validation contains only ~1767 observations and each test (validation) set contains ~196 samples.” The correct version states “~1,479” in place of “~1767” and “~164” in place of “~196”. (Table presented.). (Figure presented.).

Original languageEnglish (US)
Article number3643
JournalNature communications
Volume11
Issue number1
DOIs
StatePublished - Dec 1 2020

ASJC Scopus subject areas

  • Chemistry(all)
  • Biochemistry, Genetics and Molecular Biology(all)
  • Physics and Astronomy(all)

Fingerprint

Dive into the research topics of 'Author Correction: Enhancing materials property prediction by leveraging computational and experimental data using deep transfer learning (Nature Communications, (2019), 10, 1, (5316), 10.1038/s41467-019-13297-w)'. Together they form a unique fingerprint.

Cite this