Certifying ensembles: a general certification theory with s-lipschitzness
Improving and guaranteeing the robustness of deep learning models has been a topic of intense research. Ensembling, which combines several classifiers to provide a better model, has shown to be beneficial for generalisation, uncertainty estimation, calibration, and mitigating the effects of concept...
| Main Authors: | , , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Journal of Machine Learning Research
2023
|
| _version_ | 1797110904874074112 |
|---|---|
| author | Petrov, A Eiras, F Sanyal, A Torr, PHS Bibi, A |
| author_facet | Petrov, A Eiras, F Sanyal, A Torr, PHS Bibi, A |
| author_sort | Petrov, A |
| collection | OXFORD |
| description | Improving and guaranteeing the robustness of
deep learning models has been a topic of intense
research. Ensembling, which combines several
classifiers to provide a better model, has shown to
be beneficial for generalisation, uncertainty estimation, calibration, and mitigating the effects of
concept drift. However, the impact of ensembling
on certified robustness is less well understood. In
this work, we generalise Lipschitz continuity by
introducing S-Lipschitz classifiers, which we use
to analyse the theoretical robustness of ensembles.
Our results are precise conditions when ensembles of robust classifiers are more robust than any
constituent classifier, as well as conditions when
they are less robust. |
| first_indexed | 2024-03-07T08:02:51Z |
| format | Conference item |
| id | oxford-uuid:eb00af37-0baa-43da-a278-73ad22fda6f2 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T08:02:51Z |
| publishDate | 2023 |
| publisher | Journal of Machine Learning Research |
| record_format | dspace |
| spelling | oxford-uuid:eb00af37-0baa-43da-a278-73ad22fda6f22023-10-02T07:54:20ZCertifying ensembles: a general certification theory with s-lipschitznessConference itemhttp://purl.org/coar/resource_type/c_5794uuid:eb00af37-0baa-43da-a278-73ad22fda6f2EnglishSymplectic ElementsJournal of Machine Learning Research2023Petrov, AEiras, FSanyal, ATorr, PHSBibi, AImproving and guaranteeing the robustness of deep learning models has been a topic of intense research. Ensembling, which combines several classifiers to provide a better model, has shown to be beneficial for generalisation, uncertainty estimation, calibration, and mitigating the effects of concept drift. However, the impact of ensembling on certified robustness is less well understood. In this work, we generalise Lipschitz continuity by introducing S-Lipschitz classifiers, which we use to analyse the theoretical robustness of ensembles. Our results are precise conditions when ensembles of robust classifiers are more robust than any constituent classifier, as well as conditions when they are less robust. |
| spellingShingle | Petrov, A Eiras, F Sanyal, A Torr, PHS Bibi, A Certifying ensembles: a general certification theory with s-lipschitzness |
| title | Certifying ensembles: a general certification theory with s-lipschitzness |
| title_full | Certifying ensembles: a general certification theory with s-lipschitzness |
| title_fullStr | Certifying ensembles: a general certification theory with s-lipschitzness |
| title_full_unstemmed | Certifying ensembles: a general certification theory with s-lipschitzness |
| title_short | Certifying ensembles: a general certification theory with s-lipschitzness |
| title_sort | certifying ensembles a general certification theory with s lipschitzness |
| work_keys_str_mv | AT petrova certifyingensemblesageneralcertificationtheorywithslipschitzness AT eirasf certifyingensemblesageneralcertificationtheorywithslipschitzness AT sanyala certifyingensemblesageneralcertificationtheorywithslipschitzness AT torrphs certifyingensemblesageneralcertificationtheorywithslipschitzness AT bibia certifyingensemblesageneralcertificationtheorywithslipschitzness |