Nair, Deebul und Hochgeschwender, Nico und Olivares-Mendez, Miguel (2022) Maximum Likelihood Uncertainty Estimation: Robustness to Outliers. In: 2022 Workshop on Artificial Intelligence Safety, SafeAI 2022. Workshop on Artificial Intelligence Safety. The Thirty-Sixth AAAI Conference on Artificial Intelligence (AAAI-22)., 2022-02-28 - 2022-03-01, Vancouver, Canada. ISSN 1613-0073.
PDF
2MB |
Kurzfassung
We benchmark the robustness of maximum likelihood based uncertainty estimation methods to outliers in training data for regression tasks. Outliers or noisy labels in training data results in degraded performances as well as incorrect estimation of uncertainty. We propose the use of a heavy-tailed distribution (Laplace distribution) to improve the robustness to outliers. This property is evaluated using standard regression benchmarks and on a high-dimensional regression task of monocular depth estimation, both containing outliers. In particular, heavy-tailed distribution based maximum likelihood provides better uncertainty estimates, better separation in uncertainty for out-of-distribution data, as well as better detection of adversarial attacks in the presence of outliers.
elib-URL des Eintrags: | https://elib.dlr.de/191835/ | ||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Dokumentart: | Konferenzbeitrag (Vortrag) | ||||||||||||||||
Titel: | Maximum Likelihood Uncertainty Estimation: Robustness to Outliers | ||||||||||||||||
Autoren: |
| ||||||||||||||||
Datum: | März 2022 | ||||||||||||||||
Erschienen in: | 2022 Workshop on Artificial Intelligence Safety, SafeAI 2022 | ||||||||||||||||
Referierte Publikation: | Ja | ||||||||||||||||
Open Access: | Ja | ||||||||||||||||
Gold Open Access: | Nein | ||||||||||||||||
In SCOPUS: | Ja | ||||||||||||||||
In ISI Web of Science: | Nein | ||||||||||||||||
ISSN: | 1613-0073 | ||||||||||||||||
Status: | veröffentlicht | ||||||||||||||||
Stichwörter: | Deep Learning, Machine Learning, Robustness, Uncertainty Estimation | ||||||||||||||||
Veranstaltungstitel: | Workshop on Artificial Intelligence Safety. The Thirty-Sixth AAAI Conference on Artificial Intelligence (AAAI-22). | ||||||||||||||||
Veranstaltungsort: | Vancouver, Canada | ||||||||||||||||
Veranstaltungsart: | Workshop | ||||||||||||||||
Veranstaltungsbeginn: | 28 Februar 2022 | ||||||||||||||||
Veranstaltungsende: | 1 März 2022 | ||||||||||||||||
HGF - Forschungsbereich: | Luftfahrt, Raumfahrt und Verkehr | ||||||||||||||||
HGF - Programm: | Raumfahrt | ||||||||||||||||
HGF - Programmthema: | Technik für Raumfahrtsysteme | ||||||||||||||||
DLR - Schwerpunkt: | Raumfahrt | ||||||||||||||||
DLR - Forschungsgebiet: | R SY - Technik für Raumfahrtsysteme | ||||||||||||||||
DLR - Teilgebiet (Projekt, Vorhaben): | R - Cognitive Autonomy for Space Systems (CASSy) | ||||||||||||||||
Standort: | Köln-Porz | ||||||||||||||||
Institute & Einrichtungen: | Institut für Simulations- und Softwaretechnik Institut für Simulations- und Softwaretechnik > Verteilte Systeme und Komponentensoftware | ||||||||||||||||
Hinterlegt von: | Hochgeschwender, Nico | ||||||||||||||||
Hinterlegt am: | 22 Dez 2022 08:32 | ||||||||||||||||
Letzte Änderung: | 13 Nov 2024 15:17 |
Nur für Mitarbeiter des Archivs: Kontrollseite des Eintrags