000897359 001__ 897359
000897359 005__ 20240507205536.0
000897359 0247_ $$2doi$$a10.3389/frwa.2021.745563
000897359 0247_ $$2Handle$$a2128/28759
000897359 0247_ $$2altmetric$$aaltmetric:113500999
000897359 0247_ $$2WOS$$aWOS:000705052600001
000897359 037__ $$aFZJ-2021-03747
000897359 082__ $$a333.7
000897359 1001_ $$0P:(DE-Juel1)178989$$aTesch, Tobias$$b0$$eCorresponding author$$ufzj
000897359 245__ $$aVariant Approach for Identifying Spurious Relations that Deep Learning Models Learn
000897359 260__ $$aLausanne$$bFrontiers Media$$c2021
000897359 3367_ $$2DRIVER$$aarticle
000897359 3367_ $$2DataCite$$aOutput Types/Journal article
000897359 3367_ $$0PUB:(DE-HGF)16$$2PUB:(DE-HGF)$$aJournal Article$$bjournal$$mjournal$$s1715083831_7023
000897359 3367_ $$2BibTeX$$aARTICLE
000897359 3367_ $$2ORCID$$aJOURNAL_ARTICLE
000897359 3367_ $$00$$2EndNote$$aJournal Article
000897359 520__ $$aA deep learning (DL) model learns a function relating a set of input variables with a set of target variables. While the representation of this function in form of the DL model often lacks interpretability, several interpretation methods exist that provide descriptions of the function (e.g., measures of feature importance). On the one hand, these descriptions may build trust in the model or reveal its limitations. On the other hand, they may lead to new scientific understanding. In any case, a description is only useful if one is able to identify if parts of it reflect spurious instead of causal relations (e.g., random associations in the training data instead of associations due to a physical process). However, this can be challenging even for experts because, in scientific tasks, causal relations between input and target variables are often unknown or extremely complex. Commonly, this challenge is addressed by training separate instances of the considered model on random samples of the training set and identifying differences between the obtained descriptions. Here, we demonstrate that this may not be sufficient and propose to additionally consider more general modifications of the prediction task. We refer to the proposed approach as variant approach and demonstrate its usefulness and its superiority over pure sampling approaches with two illustrative prediction tasks from hydrometeorology. While being conceptually simple, to our knowledge the approach has not been formalized and systematically evaluated before.
000897359 536__ $$0G:(DE-HGF)POF4-2173$$a2173 - Agro-biogeosystems: controls, feedbacks and impact (POF4-217)$$cPOF4-217$$fPOF IV$$x0
000897359 588__ $$aDataset connected to CrossRef, Journals: juser.fz-juelich.de
000897359 7001_ $$0P:(DE-Juel1)151405$$aKollet, Stefan$$b1$$ufzj
000897359 7001_ $$0P:(DE-HGF)0$$aGarcke, Jochen$$b2
000897359 773__ $$0PERI:(DE-600)2986721-6$$a10.3389/frwa.2021.745563$$gVol. 3, p. 745563$$p745563$$tFrontiers in water$$v3$$x2624-9375$$y2021
000897359 8564_ $$uhttps://juser.fz-juelich.de/record/897359/files/Tesch%20et%20al%202021%20Variant%20Approach%20for%20Identifying%20Spurious%20Relations%20that%20Deep%20Learning%20Models%20Learn.pdf$$yOpenAccess
000897359 909CO $$ooai:juser.fz-juelich.de:897359$$pdnbdelivery$$pdriver$$popenaire$$pVDB$$popen_access
000897359 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)178989$$aForschungszentrum Jülich$$b0$$kFZJ
000897359 9101_ $$0I:(DE-588b)5008462-8$$6P:(DE-Juel1)151405$$aForschungszentrum Jülich$$b1$$kFZJ
000897359 9131_ $$0G:(DE-HGF)POF4-217$$1G:(DE-HGF)POF4-210$$2G:(DE-HGF)POF4-200$$3G:(DE-HGF)POF4$$4G:(DE-HGF)POF$$9G:(DE-HGF)POF4-2173$$aDE-HGF$$bForschungsbereich Erde und Umwelt$$lErde im Wandel – Unsere Zukunft nachhaltig gestalten$$vFür eine nachhaltige Bio-Ökonomie – von Ressourcen zu Produkten$$x0
000897359 9141_ $$y2021
000897359 915__ $$0LIC:(DE-HGF)CCBY4$$2HGFVOC$$aCreative Commons Attribution CC BY 4.0
000897359 915__ $$0StatID:(DE-HGF)0501$$2StatID$$aDBCoverage$$bDOAJ Seal$$d2020-09-06
000897359 915__ $$0StatID:(DE-HGF)0500$$2StatID$$aDBCoverage$$bDOAJ$$d2020-09-06
000897359 915__ $$0StatID:(DE-HGF)0510$$2StatID$$aOpenAccess
000897359 915__ $$0StatID:(DE-HGF)0030$$2StatID$$aPeer Review$$bDOAJ : Blind peer review$$d2020-09-06
000897359 915__ $$0StatID:(DE-HGF)0561$$2StatID$$aArticle Processing Charges$$d2020-09-06
000897359 915__ $$0StatID:(DE-HGF)0700$$2StatID$$aFees$$d2020-09-06
000897359 915__ $$0StatID:(DE-HGF)0100$$2StatID$$aJCR$$bFRONT WATER : 2022$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)0200$$2StatID$$aDBCoverage$$bSCOPUS$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)0300$$2StatID$$aDBCoverage$$bMedline$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)0501$$2StatID$$aDBCoverage$$bDOAJ Seal$$d2021-05-03T10:51:43Z
000897359 915__ $$0StatID:(DE-HGF)0500$$2StatID$$aDBCoverage$$bDOAJ$$d2021-05-03T10:51:43Z
000897359 915__ $$0StatID:(DE-HGF)0030$$2StatID$$aPeer Review$$bDOAJ : Anonymous peer review$$d2021-05-03T10:51:43Z
000897359 915__ $$0LIC:(DE-HGF)CCBYNV$$2V:(DE-HGF)$$aCreative Commons Attribution CC BY (No Version)$$bDOAJ$$d2021-05-03T10:51:43Z
000897359 915__ $$0StatID:(DE-HGF)0199$$2StatID$$aDBCoverage$$bClarivate Analytics Master Journal List$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)1050$$2StatID$$aDBCoverage$$bBIOSIS Previews$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)1190$$2StatID$$aDBCoverage$$bBiological Abstracts$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)0112$$2StatID$$aWoS$$bEmerging Sources Citation Index$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)0150$$2StatID$$aDBCoverage$$bWeb of Science Core Collection$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)1040$$2StatID$$aDBCoverage$$bZoological Record$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)9900$$2StatID$$aIF < 5$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)0561$$2StatID$$aArticle Processing Charges$$d2023-10-27
000897359 915__ $$0StatID:(DE-HGF)0700$$2StatID$$aFees$$d2023-10-27
000897359 920__ $$lyes
000897359 9201_ $$0I:(DE-Juel1)IBG-3-20101118$$kIBG-3$$lAgrosphäre$$x0
000897359 980__ $$ajournal
000897359 980__ $$aVDB
000897359 980__ $$aI:(DE-Juel1)IBG-3-20101118
000897359 980__ $$aUNRESTRICTED
000897359 9801_ $$aFullTexts