東華大學圖書館 |

Breaking the Privacy-Communication-Accuracy Trilemma in Federated Learning and Analytics /

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Breaking the Privacy-Communication-Accuracy Trilemma in Federated Learning and Analytics // Wei-Ning Chen.
作者:	Chen, Wei-Ning,
面頁冊數:	1 electronic resource (278 pages)
附註:	Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
Contained By:	Dissertations Abstracts International86-05B.
標題:	Personal information. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31652070
ISBN:	9798346382690

Breaking the Privacy-Communication-Accuracy Trilemma in Federated Learning and Analytics /
Chen, Wei-Ning,

Breaking the Privacy-Communication-Accuracy Trilemma in Federated Learning and Analytics /Wei-Ning Chen. - 1 electronic resource (278 pages)

Source: Dissertations Abstracts International, Volume: 86-05, Section: B.

Recent advancements in machine learning are largely fueled by vast quantities of high-quality data, much of which is generated by edge devices such as smartphones capturing photos and videos, or social networks hosting user-generated content. However, the traditional approach of centralizing this data for analysis faces increasing challenges due to network bandwidth limitations and the growing concerns surrounding legal, administrative, and ethical constraints in sharing sensitive or personal information. Federated learning and analytics have emerged as promising alternatives, enabling the application of data science and machine learning methods to raw data stored locally on users' devices. This framework ensures that raw data never leaves its source; instead, users communicate narrowly scoped messages to a central server to facilitate the completion of the targeted task, while ideally revealing little information about the actual data and its sensitive characteristics.The successful deployment of federated learning and analytics systems relies on three critical performance objectives:• Accuracy: the federated machine learning and analytics methods should produce estimates, models, and predictions that approach the accuracy achieved by centralized data analysis.• Privacy: these methods must yield formal guarantees for preserving the privacy of the user data since communicated messages can leak information about the underlying user data.• Communication-efficiency: the messages communicated by the users should be as succinct as possible to prevent incurring large delays and costs over bandwidth-limited networks.This thesis explores the fundamental trade-off between privacy, communication, and accuracy under various privacy and communication models. At first glance, privacy and communication may seem to be in conflict: achieving privacy requires the addition of noise, thereby increasing the entropy of the data and making it less compressible. Surprisingly, we show that these two objectives, communication efficiency and privacy, can be aligned under various popular privacy frameworks for federated learning and analytics, including central, local, and distributed DP models, so that one requirement seamlessly enhances the other. We demonstrate that with more stringent privacy requirements, we can compress more aggressively and vice versa, while maintaining the same level iv of accuracy for the final task. Below we highlight some of the key contributions we make in each chapter of this thesis.1. In Chapter 2, we characterize the exact trade-off between accuracy, privacy, and communication in federated learning under the central DP model, which assumes that the server is trusted. We demonstrate that the inherent randomness in compression can often be leveraged to "amplify" privacy guarantees.2. In Chapter 3, we characterize the order-optimal trade-off between accuracy, privacy, and communication in federated learning under the local DP model, which does not require the server to be trusted. We propose several schemes for distributed mean estimation that achieve the three-way optimal trade-off order-wise or exactly.3. In Chapter 4, we introduce a distributed DP framework incorporating multi-party computation (MPC), enabling us to achieve the three-way optimal trade-off under the central model without a trusted server. We develop communication-efficient schemes for this setting and prove that this model incurs unavoidable additional communication costs compared to central DP.4. In Chapter 5 and 6, we explore the frequency estimation problem for federated analytics.

English

ISBN: 9798346382690Subjects--Topical Terms:

3562412
Personal information.

Breaking the Privacy-Communication-Accuracy Trilemma in Federated Learning and Analytics /
LDR:04966nmm a22003973i 4500 001 2434183
005 20250912100610.5
006 m o d
007 cr|nu||||||||
008 261011s2024 miu||||||m |||||||eng d
020 $a 9798346382690
035 $a (MiAaPQD)AAI31652070
035 $a (MiAaPQD)Stanfordzv948ts2688
035 $a AAI31652070
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Chen, Wei-Ning, $e author. $3 3824077
245 1 0 $a Breaking the Privacy-Communication-Accuracy Trilemma in Federated Learning and Analytics / $c Wei-Ning Chen.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2024
300 $a 1 electronic resource (278 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-05, Section: B.
500 $a Advisors: Özgür, Ayfer Committee members: Pilanci, Mert; Weissman, Tsachy.
502 $b Ph.D. $c Stanford University $d 2024.
520 $a Recent advancements in machine learning are largely fueled by vast quantities of high-quality data, much of which is generated by edge devices such as smartphones capturing photos and videos, or social networks hosting user-generated content. However, the traditional approach of centralizing this data for analysis faces increasing challenges due to network bandwidth limitations and the growing concerns surrounding legal, administrative, and ethical constraints in sharing sensitive or personal information. Federated learning and analytics have emerged as promising alternatives, enabling the application of data science and machine learning methods to raw data stored locally on users' devices. This framework ensures that raw data never leaves its source; instead, users communicate narrowly scoped messages to a central server to facilitate the completion of the targeted task, while ideally revealing little information about the actual data and its sensitive characteristics.The successful deployment of federated learning and analytics systems relies on three critical performance objectives:• Accuracy: the federated machine learning and analytics methods should produce estimates, models, and predictions that approach the accuracy achieved by centralized data analysis.• Privacy: these methods must yield formal guarantees for preserving the privacy of the user data since communicated messages can leak information about the underlying user data.• Communication-efficiency: the messages communicated by the users should be as succinct as possible to prevent incurring large delays and costs over bandwidth-limited networks.This thesis explores the fundamental trade-off between privacy, communication, and accuracy under various privacy and communication models. At first glance, privacy and communication may seem to be in conflict: achieving privacy requires the addition of noise, thereby increasing the entropy of the data and making it less compressible. Surprisingly, we show that these two objectives, communication efficiency and privacy, can be aligned under various popular privacy frameworks for federated learning and analytics, including central, local, and distributed DP models, so that one requirement seamlessly enhances the other. We demonstrate that with more stringent privacy requirements, we can compress more aggressively and vice versa, while maintaining the same level iv of accuracy for the final task. Below we highlight some of the key contributions we make in each chapter of this thesis.1. In Chapter 2, we characterize the exact trade-off between accuracy, privacy, and communication in federated learning under the central DP model, which assumes that the server is trusted. We demonstrate that the inherent randomness in compression can often be leveraged to "amplify" privacy guarantees.2. In Chapter 3, we characterize the order-optimal trade-off between accuracy, privacy, and communication in federated learning under the local DP model, which does not require the server to be trusted. We propose several schemes for distributed mean estimation that achieve the three-way optimal trade-off order-wise or exactly.3. In Chapter 4, we introduce a distributed DP framework incorporating multi-party computation (MPC), enabling us to achieve the three-way optimal trade-off under the central model without a trusted server. We develop communication-efficient schemes for this setting and prove that this model incurs unavoidable additional communication costs compared to central DP.4. In Chapter 5 and 6, we explore the frequency estimation problem for federated analytics.
546 $a English
590 $a School code: 0212
650 4 $a Personal information. $3 3562412
650 4 $a Protocol. $3 3564384
650 4 $a Communication. $3 524709
650 4 $a Bandwidths. $3 3560998
650 4 $a User generated content. $3 3562474
650 4 $a Design. $3 518875
650 4 $a Privacy. $3 528582
650 4 $a Web studies. $3 2122754
690 $a 0389
690 $a 0800
690 $a 0459
690 $a 0646
710 2 $a Stanford University. $e degree granting institution. $3 3765820
720 1 $a Özgür, Ayfer $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-05B.
790 $a 0212
791 $a Ph.D.
792 $a 2024
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31652070