Gast ::
| Hauptseite > Publikationsdatenbank > JuHarmonize: Leakage-free data harmonization |
| Hauptseite > Publikationsdatenbank > JuHarmonize: Leakage-free data harmonization |
| Poster (After Call) | FZJ-2023-03220 |
; ;
2023
This record in other databases:
Please use a persistent id in citations: doi:10.34734/FZJ-2023-03220
Abstract: Combining datasets is desirable when building machine learning models. Differences in data acquisition present undesired variability undermining subsequent machine learning performance. Data harmonization methods such as ComBat can be employed, however, the requirement of test set labels causes data leakage and prevents real-world deployment. We propose a method called JuHarmonize that harmonizes data without those issues.
|
The record appears in these collections: |
PDF