Toward efficient resource utilization at edge nodes in federated learning DOI Creative Commons
Sadi Alawadi, Addi Ait‐Mlouk, Salman Toor

et al.

arXiv (Cornell University), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Federated learning (FL) enables edge nodes to collaboratively contribute constructing a global model without sharing their data. This is accomplished by devices computing local, private updates that are then aggregated server. However, computational resource constraints and network communication can become severe bottleneck for larger sizes typical deep applications. Edge tend have limited hardware resources (RAM, CPU), the bandwidth reliability at concern scaling federated fleet In this paper, we propose evaluate FL strategy inspired transfer in order reduce utilization on devices, as well load server each training round. For local update, randomly select layers train, freezing remaining part of model. doing so, both costs per round excluding all untrained layer weights from being transferred The goal study empirically explore potential trade-off between convergence under proposed strategy. We implement approach using framework FEDn. A number experiments were carried out over different datasets (CIFAR-10, CASA, IMDB), performing tasks deep-learning architectures. Our results show partially accelerate process, efficiently utilizes on-device, data transmission around 75% 53% when train 25%, 50% layers, respectively, harming resulting accuracy.

Language: Английский

Ensemble methods with feature selection and data balancing for improved code smells classification performance DOI Creative Commons
Pravin Singh Yadav,

Rajwant Singh Rao,

Alok Mishra

et al.

Engineering Applications of Artificial Intelligence, Journal Year: 2024, Volume and Issue: 139, P. 109527 - 109527

Published: Oct. 28, 2024

Language: Английский

Citations

4
Handling Non-IID Data in Federated Learning: An Experimental Evaluation Towards Unified Metrics DOI

M. Haller,

Christian Lenz,

R. Nachtigall

et al.

2021 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech), Journal Year: 2023, Volume and Issue: unknown, P. 0762 - 0770

Published: Nov. 14, 2023

Recent research has demonstrated that Non-Identically Distributed (Non-IID) data can negatively impact the performance of global models constructed in federated learning. To address this concern, multiple approaches have been developed. Nonetheless, previous lacks a cohesive overview and fails to uniformly assess these strategies, resulting challenges when comparing choosing relevant options for real-world scenarios. This study presents structured survey cutting-edge techniques handling Non-IID data, accompanied by proposing metric develop standardized approach assessing skew its harmony with appropriate approach. The findings affirm metric's suitability as heuristic distributed datasets without having insight into client serving both scientific practical purposes thus supporting selection strategies. preliminary establishes foundation discussing standardizing methodologies evaluating heterogeneity

Language: Английский

Citations

7
Toward efficient resource utilization at edge nodes in federated learning DOI Creative Commons
Sadi Alawadi, Addi Ait‐Mlouk, Salman Toor

et al.

Progress in Artificial Intelligence, Journal Year: 2024, Volume and Issue: 13(2), P. 101 - 117

Published: June 1, 2024

Language: Английский

Citations

2
Exploring the role of project status information in effective code smell detection DOI Creative Commons
Khalid Alkharabsheh, Sadi Alawadi, Yania Crespo

et al.

Cluster Computing, Journal Year: 2024, Volume and Issue: 28(1)

Published: Oct. 22, 2024

Abstract Repairing code smells detected in the or design of system is one activities contributing to increasing software quality. In this study, we investigate impact non-numerical information software, such as project status combined with machine learning techniques, on improving smell detection. For purpose, constructed a dataset consisting 22 systems various statuses, 12,040 classes, and 18 features that included 1935 large classes. A set experiments was conducted ten different techniques by dividing into training, validation, testing sets detect class smell. Feature selection data balancing have been applied. The classifier’s performance evaluated using six indicators: precision, recall, F-measure, MCC, ROC area, Kappa tests. preliminary experimental results reveal feature poor influence accuracy classifiers. Moreover, they vary their behavior when utilized values for selected average value classifiers fed better than without. Random Forest achieved best according all indicators (100%) information, while AdaBoostM1 SMO worst most them (> 86%). According findings providing about classes be analyzed can improve

Language: Английский

Citations

1
Toward efficient resource utilization at edge nodes in federated learning DOI Creative Commons
Sadi Alawadi, Addi Ait‐Mlouk, Salman Toor

et al.

arXiv (Cornell University), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Federated learning (FL) enables edge nodes to collaboratively contribute constructing a global model without sharing their data. This is accomplished by devices computing local, private updates that are then aggregated server. However, computational resource constraints and network communication can become severe bottleneck for larger sizes typical deep applications. Edge tend have limited hardware resources (RAM, CPU), the bandwidth reliability at concern scaling federated fleet In this paper, we propose evaluate FL strategy inspired transfer in order reduce utilization on devices, as well load server each training round. For local update, randomly select layers train, freezing remaining part of model. doing so, both costs per round excluding all untrained layer weights from being transferred The goal study empirically explore potential trade-off between convergence under proposed strategy. We implement approach using framework FEDn. A number experiments were carried out over different datasets (CIFAR-10, CASA, IMDB), performing tasks deep-learning architectures. Our results show partially accelerate process, efficiently utilizes on-device, data transmission around 75% 53% when train 25%, 50% layers, respectively, harming resulting accuracy.

Language: Английский

Citations

0