Academic Journal

Blockchain controlled trustworthy federated learning platform for smart homes

التفاصيل البيبلوغرافية
العنوان: Blockchain controlled trustworthy federated learning platform for smart homes
المؤلفون: Sujit Biswas, Kashif Sharif, Zohaib Latif, Mohammed J. F. Alenazi, Ashok Kumar Pradhan, Anupam Kumar Bairagi
المصدر: IET Communications, Vol 18, Iss 20, Pp 1840-1852 (2024)
بيانات النشر: Wiley, 2024.
سنة النشر: 2024
المجموعة: LCC:Telecommunication
مصطلحات موضوعية: computer network security, blockchain, federated learning, Telecommunication, TK5101-6720
الوصف: Abstract Smart device manufacturers rely on insights from smart home (SH) data to update their devices, and similarly, service providers use it for predictive maintenance. In terms of data security and privacy, combining distributed federated learning (FL) with blockchain technology is being considered to prevent single point failure and model poising attacks. However, adding blockchain to a FL environment can worsen blockchain's scaling issues and create regular service interruptions at SH. This article presents a scalable Blockchain‐based Privacy‐preserving Federated Learning (BPFL) architecture for an SH ecosystem that integrates blockchain and FL. BPFL can automate SHs' services and distribute machine learning (ML) operations to update IoT manufacturer models and scale service provider services. The architecture uses a local peer as a gateway to connect SHs to the blockchain network and safeguard user data, transactions, and ML operations. Blockchain facilitates ecosystem access management and learning. The Stanford Cars and an IoT dataset have been used as test bed experiments, taking into account the nature of data (i.e. images and numeric). The experiments show that ledger optimisation can boost scalability by 40–60% in BCN by reducing transaction overhead by 60%. Simultaneously, it increases learning capacity by 10% compared to baseline FL techniques.
نوع الوثيقة: article
وصف الملف: electronic resource
اللغة: English
تدمد: 1751-8636
1751-8628
Relation: https://doaj.org/toc/1751-8628; https://doaj.org/toc/1751-8636
DOI: 10.1049/cmu2.12870
URL الوصول: https://doaj.org/article/1354682a1eb1438eb381dc32b3401f15
رقم الانضمام: edsdoj.1354682a1eb1438eb381dc32b3401f15
قاعدة البيانات: Directory of Open Access Journals
الوصف
تدمد:17518636
17518628
DOI:10.1049/cmu2.12870