東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

FindBook

Google Book

Amazon

博客來

Federated Machine Learning in Edge Computing.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Federated Machine Learning in Edge Computing./
作者:	Mills, Jed.
面頁冊數:	1 online resource (100 pages)
附註:	Source: Dissertations Abstracts International, Volume: 84-09, Section: A.
Contained By:	Dissertations Abstracts International84-09A.
標題:	Computers. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30397418click for full text (PQDT)
ISBN:	9798377607021

Federated Machine Learning in Edge Computing.
Mills, Jed.

Federated Machine Learning in Edge Computing. - 1 online resource (100 pages)

Source: Dissertations Abstracts International, Volume: 84-09, Section: A.

Thesis (Ph.D.)--University of Exeter (United Kingdom), 2022.

Includes bibliographical references

Machine Learning (ML) is transforming the way that computers are used to solve problems in computer vision, natural language processing, scientific modelling, and much more. The rising number of devices connected to the Internet generate huge quantities of data that can be used for ML purposes. Traditionally, organisations require user data to be uploaded to a single location (i.e., cloud datacentre) for centralised ML. However, public concerns regarding data-privacy are growing, and in some domains such as healthcare, there exist strict laws governing the access of data. The computational power and connectivity of devices at the network edge is also increasing: edge computing is a paradigm designed to move computation from the cloud to the edge to reduce latency and traffic.Federated Learning (FL) is a new and swiftly-developing field that has huge potential for privacy-preserving ML. In FL, edge devices collaboratively train a model without users sharing their personal data with any other party. However, there exist multiple challenges for designing useful FL algorithms, including: the heterogeneity of data across participating clients; the low computing power, intermittent connectivity and unreliability of clients at the network edge compared to the datacentre; and the difficulty of limiting information leakage whilst still training high-performance models. This thesis proposes new methods for improving the process of FL in edge computing and hence making it more practical for real-world deployments. First, a novel approach is designed that accelerates the convergence of the FL model through adaptive optimisation, reducing the time taken to train a model, whilst lowering the total quantity of information uploaded from edge clients to the coordinating server through two new compression strategies. Next, a Multi-Task FL framework is proposed that allows participating clients to train unique models that are tailored to their own heterogeneous datasets whilst still benefiting from FL, improving model convergence speed and generalisation performance across clients. Then, the principle of decreasing the total work that clients perform during the FL process is explored. A theoretical analysis (and subsequent experimental evaluation) suggests that this approach can reduce the time taken to reach a desired training error whilst lowering the total computational cost of FL and improving communication-efficiency. Lastly, an algorithm is designed that applies adaptive optimisation to FL in a novel way, through the use of a statistically-biased optimiser whose values are kept fixed on clients. This algorithm can leverage the convergence guarantees of centralised algorithms, with the addition of FL-related error-terms. Furthermore, it shows excellent performance on benchmark FL datasets whilst possessing lower computation and upload costs compared to competing adaptive-FL algorithms.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2023

Mode of access: World Wide Web

ISBN: 9798377607021Subjects--Topical Terms:

544777
Computers.
Subjects--Index Terms:

Machine learningIndex Terms--Genre/Form:

542853
Electronic books.

Federated Machine Learning in Edge Computing.
LDR:04288nmm a2200397K 4500 001 2364796
005 20231212064423.5
006 m o d
007 cr mn ---uuuuu
008 241011s2022 xx obm 000 0 eng d
020 $a 9798377607021
035 $a (MiAaPQ)AAI30397418
035 $a (MiAaPQ)Exeter10871132445
035 $a AAI30397418
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Mills, Jed. $3 3705616
245 1 0 $a Federated Machine Learning in Edge Computing.
264 0 $c 2022
300 $a 1 online resource (100 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: 84-09, Section: A.
500 $a Advisor: Hu, Jia.
502 $a Thesis (Ph.D.)--University of Exeter (United Kingdom), 2022.
504 $a Includes bibliographical references
520 $a Machine Learning (ML) is transforming the way that computers are used to solve problems in computer vision, natural language processing, scientific modelling, and much more. The rising number of devices connected to the Internet generate huge quantities of data that can be used for ML purposes. Traditionally, organisations require user data to be uploaded to a single location (i.e., cloud datacentre) for centralised ML. However, public concerns regarding data-privacy are growing, and in some domains such as healthcare, there exist strict laws governing the access of data. The computational power and connectivity of devices at the network edge is also increasing: edge computing is a paradigm designed to move computation from the cloud to the edge to reduce latency and traffic.Federated Learning (FL) is a new and swiftly-developing field that has huge potential for privacy-preserving ML. In FL, edge devices collaboratively train a model without users sharing their personal data with any other party. However, there exist multiple challenges for designing useful FL algorithms, including: the heterogeneity of data across participating clients; the low computing power, intermittent connectivity and unreliability of clients at the network edge compared to the datacentre; and the difficulty of limiting information leakage whilst still training high-performance models. This thesis proposes new methods for improving the process of FL in edge computing and hence making it more practical for real-world deployments. First, a novel approach is designed that accelerates the convergence of the FL model through adaptive optimisation, reducing the time taken to train a model, whilst lowering the total quantity of information uploaded from edge clients to the coordinating server through two new compression strategies. Next, a Multi-Task FL framework is proposed that allows participating clients to train unique models that are tailored to their own heterogeneous datasets whilst still benefiting from FL, improving model convergence speed and generalisation performance across clients. Then, the principle of decreasing the total work that clients perform during the FL process is explored. A theoretical analysis (and subsequent experimental evaluation) suggests that this approach can reduce the time taken to reach a desired training error whilst lowering the total computational cost of FL and improving communication-efficiency. Lastly, an algorithm is designed that applies adaptive optimisation to FL in a novel way, through the use of a statistically-biased optimiser whose values are kept fixed on clients. This algorithm can leverage the convergence guarantees of centralised algorithms, with the addition of FL-related error-terms. Furthermore, it shows excellent performance on benchmark FL datasets whilst possessing lower computation and upload costs compared to competing adaptive-FL algorithms.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2023
538 $a Mode of access: World Wide Web
650 4 $a Computers. $3 544777
650 4 $a Privacy. $3 528582
650 4 $a User behavior. $3 3564661
650 4 $a Personal computers. $3 3561037
650 4 $a Neural networks. $3 677449
650 4 $a Computer science. $3 523869
650 4 $a Web studies. $3 2122754
653 $a Machine learning
653 $a Federated learning
653 $a Neural networks
653 $a Low computing power
655 7 $a Electronic books. $2 lcsh $3 542853
690 $a 0800
690 $a 0984
690 $a 0646
710 2 $a ProQuest Information and Learning Co. $3 783688
710 2 $a University of Exeter (United Kingdom). $3 1273400
773 0 $t Dissertations Abstracts International $g 84-09A.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30397418 $z click for full text (PQDT)