A Cautionary Tale of Decorrelating Theory Uncertainties
arXiv:2109.08159 · doi:10.1140/epjc/s10052-022-10012-w
Abstract
A variety of techniques have been proposed to train machine learning classifiers that are independent of a given feature. While this can be an essential technique for enabling background estimation, it may also be useful for reducing uncertainties. We carefully examine theory uncertainties, which typically do not have a statistical origin. We will provide explicit examples of two-point (fragmentation modeling) and continuous (higher-order corrections) uncertainties where decorrelating significantly reduces the apparent uncertainty while the actual uncertainty is much larger. These results suggest that caution should be taken when using decorrelation for these types of uncertainties as long as we do not have a complete decomposition into statistically meaningful components.
19 pages, 8 figures. v2: added to acknowledgements
References in corpus (12)
- PYTHIA 6.4 Physics and Manual
- The automated computation of tree-level and next-to-leading order differential cross sections, and their matching to parton shower simulations
- Measuring and testing dependence by correlation of distances
- Event generation with SHERPA 1.1
- Herwig++ Physics and Manual
- Jet substructure as a new Higgs search channel at the LHC
- Brownian distance covariance
- Jet substructure as a new Higgs search channel at the LHC
- Power Counting to Better Jet Observables
- Nuisance hardened data compression for fast likelihood-free inference
- Uncertainty Aware Learning for High Energy Physics
- New approaches for boosting to uniformity