Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning

Yu Zheng; Wei Song; Minxin Du; Sherman S. M. Chow; Qian Lou; Yongjun Zhao; Xiuhua Wang

doi:10.1007/978-3-031-46664-9_27

Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning

Yu Zheng, Wei Song, Minxin Du, Sherman S. M. Chow, Qian Lou, Yongjun Zhao, Xiuhua Wang

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

7 Citations (Scopus)

Abstract

Federated learning (FL) aims to derive a “better” global model without direct access to individuals’ training data. It is traditionally done by aggregation over individual gradients with differentially private (DP) noises. We study an FL variant as a new point in the privacy-performance space. Namely, cryptographic aggregation is over local models instead of gradients; each contributor then locally trains their model using a DP version of Adam upon the “feedback” (e.g., fake samples from GAN – generative adversarial networks) derived from the securely-aggregated global model. Intuitively, this achieves the best of both worlds – more “expressive” models are processed in the encrypted domain instead of just gradients, without DP’s shortcoming, while heavyweight cryptography is minimized (at only the first step instead of the entire process). Practically, we showcase this new FL variant over GAN and meta-learning, for securing new data and new tasks.

Original language	English
Title of host publication	Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings
Editors	Xiaochun Yang, Bin Wang, Heru Suhartanto, Guoren Wang, Jing Jiang, Bing Li, Huaijie Zhu, Ningning Cui
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	393-407
Number of pages	15
ISBN (Print)	9783031466632
DOIs	https://doi.org/10.1007/978-3-031-46664-9_27
Publication status	Published - Aug 2023
Externally published	Yes
Event	19th International Conference on Advanced Data Mining and Applications, ADMA 2023 - Shenyang, China Duration: 21 Aug 2023 → 23 Aug 2023

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	14177 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	19th International Conference on Advanced Data Mining and Applications, ADMA 2023
Country/Territory	China
City	Shenyang
Period	21/08/23 → 23/08/23

Keywords

Cryptography
Differential privacy
Federated learning

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

More information

10.1007/978-3-031-46664-9_27

Cite this

Zheng, Y., Song, W., Du, M., S. M. Chow, S., Lou, Q., Zhao, Y., & Wang, X. (2023). Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning. In X. Yang, B. Wang, H. Suhartanto, G. Wang, J. Jiang, B. Li, H. Zhu, & N. Cui (Eds.), Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings (pp. 393-407). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14177 LNAI). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-46664-9_27

Zheng, Yu ; Song, Wei ; Du, Minxin et al. / Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning. Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings. editor / Xiaochun Yang ; Bin Wang ; Heru Suhartanto ; Guoren Wang ; Jing Jiang ; Bing Li ; Huaijie Zhu ; Ningning Cui. Springer Science and Business Media Deutschland GmbH, 2023. pp. 393-407 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{ea51294a6b6a45e6b0c44c8f1ab5845a,

title = "Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning",

abstract = "Federated learning (FL) aims to derive a “better” global model without direct access to individuals{\textquoteright} training data. It is traditionally done by aggregation over individual gradients with differentially private (DP) noises. We study an FL variant as a new point in the privacy-performance space. Namely, cryptographic aggregation is over local models instead of gradients; each contributor then locally trains their model using a DP version of Adam upon the “feedback” (e.g., fake samples from GAN – generative adversarial networks) derived from the securely-aggregated global model. Intuitively, this achieves the best of both worlds – more “expressive” models are processed in the encrypted domain instead of just gradients, without DP{\textquoteright}s shortcoming, while heavyweight cryptography is minimized (at only the first step instead of the entire process). Practically, we showcase this new FL variant over GAN and meta-learning, for securing new data and new tasks.",

keywords = "Cryptography, Differential privacy, Federated learning",

author = "Yu Zheng and Wei Song and Minxin Du and {S. M. Chow}, Sherman and Qian Lou and Yongjun Zhao and Xiuhua Wang",

note = "Publisher Copyright: {\textcopyright} 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 19th International Conference on Advanced Data Mining and Applications, ADMA 2023 ; Conference date: 21-08-2023 Through 23-08-2023",

year = "2023",

month = aug,

doi = "10.1007/978-3-031-46664-9_27",

language = "English",

isbn = "9783031466632",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "393--407",

editor = "Xiaochun Yang and Bin Wang and Heru Suhartanto and Guoren Wang and Jing Jiang and Bing Li and Huaijie Zhu and Ningning Cui",

booktitle = "Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings",

address = "Germany",

}

Zheng, Y, Song, W, Du, M, S. M. Chow, S, Lou, Q, Zhao, Y & Wang, X 2023, Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning. in X Yang, B Wang, H Suhartanto, G Wang, J Jiang, B Li, H Zhu & N Cui (eds), Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14177 LNAI, Springer Science and Business Media Deutschland GmbH, pp. 393-407, 19th International Conference on Advanced Data Mining and Applications, ADMA 2023, Shenyang, China, 21/08/23. https://doi.org/10.1007/978-3-031-46664-9_27

Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning. / Zheng, Yu; Song, Wei; Du, Minxin et al.
Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings. ed. / Xiaochun Yang; Bin Wang; Heru Suhartanto; Guoren Wang; Jing Jiang; Bing Li; Huaijie Zhu; Ningning Cui. Springer Science and Business Media Deutschland GmbH, 2023. p. 393-407 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14177 LNAI).

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

TY - GEN

T1 - Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning

AU - Zheng, Yu

AU - Song, Wei

AU - Du, Minxin

AU - S. M. Chow, Sherman

AU - Lou, Qian

AU - Zhao, Yongjun

AU - Wang, Xiuhua

PY - 2023/8

Y1 - 2023/8

N2 - Federated learning (FL) aims to derive a “better” global model without direct access to individuals’ training data. It is traditionally done by aggregation over individual gradients with differentially private (DP) noises. We study an FL variant as a new point in the privacy-performance space. Namely, cryptographic aggregation is over local models instead of gradients; each contributor then locally trains their model using a DP version of Adam upon the “feedback” (e.g., fake samples from GAN – generative adversarial networks) derived from the securely-aggregated global model. Intuitively, this achieves the best of both worlds – more “expressive” models are processed in the encrypted domain instead of just gradients, without DP’s shortcoming, while heavyweight cryptography is minimized (at only the first step instead of the entire process). Practically, we showcase this new FL variant over GAN and meta-learning, for securing new data and new tasks.

AB - Federated learning (FL) aims to derive a “better” global model without direct access to individuals’ training data. It is traditionally done by aggregation over individual gradients with differentially private (DP) noises. We study an FL variant as a new point in the privacy-performance space. Namely, cryptographic aggregation is over local models instead of gradients; each contributor then locally trains their model using a DP version of Adam upon the “feedback” (e.g., fake samples from GAN – generative adversarial networks) derived from the securely-aggregated global model. Intuitively, this achieves the best of both worlds – more “expressive” models are processed in the encrypted domain instead of just gradients, without DP’s shortcoming, while heavyweight cryptography is minimized (at only the first step instead of the entire process). Practically, we showcase this new FL variant over GAN and meta-learning, for securing new data and new tasks.

KW - Cryptography

KW - Differential privacy

KW - Federated learning

UR - http://www.scopus.com/inward/record.url?scp=85177606575&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-46664-9_27

DO - 10.1007/978-3-031-46664-9_27

M3 - Conference article published in proceeding or book

AN - SCOPUS:85177606575

SN - 9783031466632

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 393

EP - 407

BT - Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings

A2 - Yang, Xiaochun

A2 - Wang, Bin

A2 - Suhartanto, Heru

A2 - Wang, Guoren

A2 - Jiang, Jing

A2 - Li, Bing

A2 - Zhu, Huaijie

A2 - Cui, Ningning

PB - Springer Science and Business Media Deutschland GmbH

T2 - 19th International Conference on Advanced Data Mining and Applications, ADMA 2023

Y2 - 21 August 2023 through 23 August 2023

ER -

Zheng Y, Song W, Du M, S. M. Chow S, Lou Q, Zhao Y et al. Cryptography-Inspired Federated Learning for Generative Adversarial Networks and Meta Learning. In Yang X, Wang B, Suhartanto H, Wang G, Jiang J, Li B, Zhu H, Cui N, editors, Advanced Data Mining and Applications - 19th International Conference, ADMA 2023, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 393-407. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-46664-9_27