Zen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning

In this study, we aim to bridge the gap between the theoretical understanding of attacks against collaborative machine learning workflows and their practical ramifications by considering the effects of model architecture, learning setting and hyperparameters on the resilience against attacks. We ref...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Usynin Dmitrii, Rueckert Daniel, Passerat-Palmbach Jonathan, Kaissis Georgios
Formato:	article
Lenguaje:	EN
Publicado:	Sciendo 2022
Materias:	privacy computer vision federated learning membership inference model inversion Ethics BJ1-1725 Electronic computers. Computer science QA75.5-76.95
Acceso en línea:	https://doaj.org/article/56cd977fda7b4e01ba8ccebbda7d6e6e
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:56cd977fda7b4e01ba8ccebbda7d6e6e
record_format	dspace
spelling	oai:doaj.org-article:56cd977fda7b4e01ba8ccebbda7d6e6e2021-12-05T14:11:10ZZen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning2299-098410.2478/popets-2022-0014https://doaj.org/article/56cd977fda7b4e01ba8ccebbda7d6e6e2022-01-01T00:00:00Zhttps://doi.org/10.2478/popets-2022-0014https://doaj.org/toc/2299-0984In this study, we aim to bridge the gap between the theoretical understanding of attacks against collaborative machine learning workflows and their practical ramifications by considering the effects of model architecture, learning setting and hyperparameters on the resilience against attacks. We refer to such mitigations as model adaptation. Through extensive experimentation on both, benchmark and real-life datasets, we establish a more practical threat model for collaborative learning scenarios. In particular, we evaluate the impact of model adaptation by implementing a range of attacks belonging to the broader categories of model inversion and membership inference. Our experiments yield two noteworthy outcomes: they demonstrate the difficulty of actually conducting successful attacks under realistic settings when model adaptation is employed and they highlight the challenge inherent in successfully combining model adaptation and formal privacy-preserving techniques to retain the optimal balance between model utility and attack resilience.Usynin DmitriiRueckert DanielPasserat-Palmbach JonathanKaissis GeorgiosSciendoarticleprivacycomputer visionfederated learningmembership inferencemodel inversionEthicsBJ1-1725Electronic computers. Computer scienceQA75.5-76.95ENProceedings on Privacy Enhancing Technologies, Vol 2022, Iss 1, Pp 274-290 (2022)
institution	DOAJ
collection	DOAJ
language	EN
topic	privacy computer vision federated learning membership inference model inversion Ethics BJ1-1725 Electronic computers. Computer science QA75.5-76.95
spellingShingle	privacy computer vision federated learning membership inference model inversion Ethics BJ1-1725 Electronic computers. Computer science QA75.5-76.95 Usynin Dmitrii Rueckert Daniel Passerat-Palmbach Jonathan Kaissis Georgios Zen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning
description	In this study, we aim to bridge the gap between the theoretical understanding of attacks against collaborative machine learning workflows and their practical ramifications by considering the effects of model architecture, learning setting and hyperparameters on the resilience against attacks. We refer to such mitigations as model adaptation. Through extensive experimentation on both, benchmark and real-life datasets, we establish a more practical threat model for collaborative learning scenarios. In particular, we evaluate the impact of model adaptation by implementing a range of attacks belonging to the broader categories of model inversion and membership inference. Our experiments yield two noteworthy outcomes: they demonstrate the difficulty of actually conducting successful attacks under realistic settings when model adaptation is employed and they highlight the challenge inherent in successfully combining model adaptation and formal privacy-preserving techniques to retain the optimal balance between model utility and attack resilience.
format	article
author	Usynin Dmitrii Rueckert Daniel Passerat-Palmbach Jonathan Kaissis Georgios
author_facet	Usynin Dmitrii Rueckert Daniel Passerat-Palmbach Jonathan Kaissis Georgios
author_sort	Usynin Dmitrii
title	Zen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning
title_short	Zen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning
title_full	Zen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning
title_fullStr	Zen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning
title_full_unstemmed	Zen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning
title_sort	zen and the art of model adaptation: low-utility-cost attack mitigations in collaborative machine learning
publisher	Sciendo
publishDate	2022
url	https://doaj.org/article/56cd977fda7b4e01ba8ccebbda7d6e6e
work_keys_str_mv	AT usynindmitrii zenandtheartofmodeladaptationlowutilitycostattackmitigationsincollaborativemachinelearning AT rueckertdaniel zenandtheartofmodeladaptationlowutilitycostattackmitigationsincollaborativemachinelearning AT passeratpalmbachjonathan zenandtheartofmodeladaptationlowutilitycostattackmitigationsincollaborativemachinelearning AT kaissisgeorgios zenandtheartofmodeladaptationlowutilitycostattackmitigationsincollaborativemachinelearning
_version_	1718371300771102720

Zen and the art of model adaptation: Low-utility-cost attack mitigations in collaborative machine learning

Ejemplares similares