Mehrtens, Hendrik Alexander ; Gonzalez, Camila ; Mukhopadhyay, Anirban (2022)
Improving Robustness and Calibration in Ensembles with Diversity Regularization.
4th DAGM German Conference on Pattern Recognition. Konstanz, Germany (27.-30.09.2022)
doi: 10.1007/978-3-031-16788-1_3
Konferenzveröffentlichung, Bibliographie
Kurzbeschreibung (Abstract)
Calibration and uncertainty estimation are crucial topics in high-risk environments. Following the recent interest in the diversity of ensembles, we systematically evaluate the viability of explicitly regularizing ensemble diversity to improve robustness and calibration on in-distribution data as well as under dataset shift. We introduce a new diversity regularizer for classification tasks that uses outof-distribution samples and increases the overall accuracy, calibration and out-of-distribution detection capabilities of ensembles. We demonstrate that diversity regularization is highly beneficial in architectures where weights are partially shared between the individual members and even allows to use fewer ensemble members to reach the same level of robustness. Experiments on CIFAR-10, CIFAR-100, and SVHN show that regularizing diversity can have a significant impact on calibration and robustness, as well as out-of-distribution detection.
| Typ des Eintrags: | Konferenzveröffentlichung |
|---|---|
| Erschienen: | 2022 |
| Autor(en): | Mehrtens, Hendrik Alexander ; Gonzalez, Camila ; Mukhopadhyay, Anirban |
| Art des Eintrags: | Bibliographie |
| Titel: | Improving Robustness and Calibration in Ensembles with Diversity Regularization |
| Sprache: | Englisch |
| Publikationsjahr: | 20 September 2022 |
| Verlag: | Springer |
| Buchtitel: | Pattern Recognition: 44th DAGM German Conference |
| Reihe: | Lecture Notes in Computer Science |
| Band einer Reihe: | 13485 |
| Veranstaltungstitel: | 4th DAGM German Conference on Pattern Recognition |
| Veranstaltungsort: | Konstanz, Germany |
| Veranstaltungsdatum: | 27.-30.09.2022 |
| DOI: | 10.1007/978-3-031-16788-1_3 |
| Kurzbeschreibung (Abstract): | Calibration and uncertainty estimation are crucial topics in high-risk environments. Following the recent interest in the diversity of ensembles, we systematically evaluate the viability of explicitly regularizing ensemble diversity to improve robustness and calibration on in-distribution data as well as under dataset shift. We introduce a new diversity regularizer for classification tasks that uses outof-distribution samples and increases the overall accuracy, calibration and out-of-distribution detection capabilities of ensembles. We demonstrate that diversity regularization is highly beneficial in architectures where weights are partially shared between the individual members and even allows to use fewer ensemble members to reach the same level of robustness. Experiments on CIFAR-10, CIFAR-100, and SVHN show that regularizing diversity can have a significant impact on calibration and robustness, as well as out-of-distribution detection. |
| Freie Schlagworte: | Diversity, Ensembles, Robustness, Calibration |
| Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik 20 Fachbereich Informatik > Graphisch-Interaktive Systeme |
| Hinterlegungsdatum: | 06 Jun 2023 14:02 |
| Letzte Änderung: | 02 Aug 2023 14:15 |
| PPN: | 510089844 |
| Export: | |
| Suche nach Titel in: | TUfind oder in Google |
 |
Frage zum Eintrag |
Optionen (nur für Redakteure)
 |
Redaktionelle Details anzeigen |
Drucken
Impressum