TUdatalib : Fast Axiomatic Attribution for Neural Networks

Zur Kurzanzeige

dc.contributor.author	Hesse, Robin
dc.contributor.author	Schaub-Meyer, Simone
dc.contributor.author	Roth, Stefan
dc.date.accessioned	2023-08-04T09:31:38Z
dc.date.available	2022-01-19T09:36:04Z
dc.date.available	2023-08-04T09:31:38Z
dc.date.issued	2021-12
dc.identifier.uri	https://tudatalib.ulb.tu-darmstadt.de/handle/tudatalib/3389.2
dc.description	Mitigating the dependence on spurious correlations present in the training dataset is a quickly emerging and important topic of deep learning. Recent approaches include priors on the feature attribution of a deep neural network (DNN) into the training process to reduce the dependence on unwanted features. However, until now one needed to trade off high-quality attributions, satisfying desirable axioms, against the time required to compute them. This in turn either led to long training times or ineffective attribution priors. In this work, we break this trade-off by considering a special class of efficiently axiomatically attributable DNNs for which an axiomatic feature attribution can be computed with only a single forward/backward pass. We formally prove that nonnegatively homogeneous DNNs, here termed X-DNNs, are efficiently axiomatically attributable and show that they can be effortlessly constructed from a wide range of regular DNNs by simply removing the bias term of each layer. Various experiments demonstrate the advantages of X-DNNs, beating state-of-the-art generic attribution methods on regular DNNs for training with attribution priors.	de_DE
dc.language.iso	en	de_DE
dc.relation	IsDescribedBy;arXiv;https://arxiv.org/abs/2111.07668
dc.rights	Apache License 2.0
dc.rights.uri	https://www.apache.org/licenses/LICENSE-2.0
dc.subject	deep learning	de_DE
dc.subject	interpretability	de_DE
dc.subject	attribution	de_DE
dc.subject	attribution prior	de_DE
dc.subject.classification	409-05 Interaktive und intelligente Systeme, Bild- und Sprachverarbeitung, Computergraphik und Visualisierung	de_DE
dc.subject.ddc	004
dc.title	Fast Axiomatic Attribution for Neural Networks	de_DE
dc.type	Software	de_DE
tud.project	EC/H2020 \| 866008 \| RED	de_DE
tud.project	HMWK \| 500/10.001-(00111) \| 3AI-NWG Schaub-Meyer	de_DE
tud.project	HMWK \| 500/10.001-(00111) \| 3AI - TP Roth	de_DE
tud.project	HMWK \| 500/10.001-(00012) \| TAM - TP Roth	de_DE
tud.unit	TUDa