Low-Cost Bayesian Methods for Fixing Neural Networks' Overconfidence

DSpace Repositorium (Manakin basiert)

Zur Kurzanzeige

dc.contributor.advisor Hennig, Philipp (Prof. Dr.)
dc.contributor.author Kristiadi, Agustinus
dc.date.accessioned 2023-01-20T10:11:15Z
dc.date.available 2023-01-20T10:11:15Z
dc.date.issued 2023-01-20
dc.identifier.uri http://hdl.handle.net/10900/135535
dc.identifier.uri http://nbn-resolving.de/urn:nbn:de:bsz:21-dspace-1355355 de_DE
dc.identifier.uri http://dx.doi.org/10.15496/publikation-76886
dc.description.abstract Well-calibrated predictive uncertainty of neural networks—essentially making them know when they do not know—is paramount in safety-critical applications. However, deep neural networks are overconfident in the region both far away and near the training data. In this thesis, we study Bayesian neural networks and their extensions to mitigate this issue. First, we show that being Bayesian, even just at the last layer and in a post-hoc manner via Laplace approximations, helps mitigate overconfidence in deep ReLU classifiers. Then, we provide a cost-effective Gaussian-process extension to ReLU Bayesian neural networks that provides a guarantee that ReLU nets will never be overconfident in the region far from the data. Furthermore, we propose three ways of improving the calibration of general Bayesian neural networks in the regions near the data by (i) refining parametric approximations to the Bayesian neural networks’ posteriors with normalizing flows, (ii) training the uncertainty of Laplace approximations, and (iii) leveraging out-of-distribution data during training. We provide an easy-to-use library, laplace-torch, to facilitate the modern arts of Laplace approximations in deep learning. It gives users a way to turn a standard pre-trained deep net into a Bayesian neural network in a cost-efficient manner. en
dc.language.iso en de_DE
dc.publisher Universität Tübingen de_DE
dc.rights ubt-podok de_DE
dc.rights.uri http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=de de_DE
dc.rights.uri http://tobias-lib.uni-tuebingen.de/doku/lic_mit_pod.php?la=en en
dc.subject.classification Maschinelles Lernen , Neuronales Netz de_DE
dc.subject.ddc 004 de_DE
dc.subject.other Neural Network en
dc.subject.other Bayesian Deep Learning en
dc.subject.other Uncertainty Quantification en
dc.subject.other Laplace Approximations en
dc.title Low-Cost Bayesian Methods for Fixing Neural Networks' Overconfidence en
dc.type PhDThesis de_DE
dcterms.dateAccepted 2023-01-13
utue.publikation.fachbereich Informatik de_DE
utue.publikation.fakultaet 7 Mathematisch-Naturwissenschaftliche Fakultät de_DE
utue.publikation.noppn yes de_DE

Dateien:

Das Dokument erscheint in:

Zur Kurzanzeige