Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Distributed client selection with multi-objective in federated learning assisted Internet of Vehicles

Narisu Cha, Long Chang

TL;DR

The paper addresses the large overhead of maintaining an active state in federated learning over IoV environments with millions of vehicles. It proposes a distributed client selection framework in which evaluation and selection are performed locally (via neighbor exchanges) using a multi-objective fuzzy evaluator that considers sample quantity, throughput, computational capability, and local dataset diversity. Key contributions include (i) a decentralized client selection mechanism that avoids collecting all participant states at a central server, (ii) a fuzzy Mamdani-based evaluator to handle four heterogeneous inputs, (iii) throughput prediction and training-time modeling to inform decision making, and (iv) extensive simulations showing the approach closely approximates centralized selection while reducing active-state overhead. The findings indicate significant reductions in communication overhead and robust performance under non-i.i.d data and mobility, enabling scalable FL in IoV with practical resource usage.

Abstract

Federated learning is an emerging distributed machine learning framework in the Internet of Vehicles (IoV). In IoV, millions of vehicles are willing to train the model to share their knowledge. Maintaining an active state means the participants must update their state to the FL server in a fixed interval and participate to next round. However, the cost by maintaining an active state is very large when there are a huge number of participating vehicles. In this paper, we proposed a distributed client selection scheme to reduce the cost of maintaining the active state for all participants. The clients with the highest evaluation are elected among the neighbours. In the evaluator, four variables are considered including sample quantity, throughput available, computational capability and the quality of the local dataset. We adopted fuzzy logic as the evaluator since the closed-form solution over four variables does not exist. Extensive simulation results show our proposal approximates the centralized client selection in terms of accuracy and can significantly reduce the communication overhead.

Distributed client selection with multi-objective in federated learning assisted Internet of Vehicles

TL;DR

The paper addresses the large overhead of maintaining an active state in federated learning over IoV environments with millions of vehicles. It proposes a distributed client selection framework in which evaluation and selection are performed locally (via neighbor exchanges) using a multi-objective fuzzy evaluator that considers sample quantity, throughput, computational capability, and local dataset diversity. Key contributions include (i) a decentralized client selection mechanism that avoids collecting all participant states at a central server, (ii) a fuzzy Mamdani-based evaluator to handle four heterogeneous inputs, (iii) throughput prediction and training-time modeling to inform decision making, and (iv) extensive simulations showing the approach closely approximates centralized selection while reducing active-state overhead. The findings indicate significant reductions in communication overhead and robust performance under non-i.i.d data and mobility, enabling scalable FL in IoV with practical resource usage.

Abstract

Federated learning is an emerging distributed machine learning framework in the Internet of Vehicles (IoV). In IoV, millions of vehicles are willing to train the model to share their knowledge. Maintaining an active state means the participants must update their state to the FL server in a fixed interval and participate to next round. However, the cost by maintaining an active state is very large when there are a huge number of participating vehicles. In this paper, we proposed a distributed client selection scheme to reduce the cost of maintaining the active state for all participants. The clients with the highest evaluation are elected among the neighbours. In the evaluator, four variables are considered including sample quantity, throughput available, computational capability and the quality of the local dataset. We adopted fuzzy logic as the evaluator since the closed-form solution over four variables does not exist. Extensive simulation results show our proposal approximates the centralized client selection in terms of accuracy and can significantly reduce the communication overhead.
Paper Structure (18 sections, 9 equations, 9 figures, 3 tables, 1 algorithm)

This paper contains 18 sections, 9 equations, 9 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Federated learning assisted IoV
  4. Federated learning
  5. Federated learning assisted IoV
  6. Distributed client selection framework
  7. Different client selection schemes
  8. Communication overhead
  9. Network model
  10. Distributed client selection framework
  11. Multi-objective evaluator
  12. Prediction to the network throughput
  13. Training Time
  14. Fuzzy evaluator with multi-objective
  15. Simulation and evaluation
  16. ...and 3 more sections

Figures (9)

  • Figure 1: Different client selection schemes in FL. (a) Client selection in the CFL. (b) Client selection in the CFL-fuzzy fuzzyCentralized2022. (c) Distributed client selection.
  • Figure 2: Comparison of two kinds of overhead. The dashed red line presents the size of the uploading model in each round. The blue and green lines present the overhead by maintaining the active state for all participants in the CFL and CFL-fuzzy, respectively.
  • Figure 3: The network architecture of distributed client selection framework.
  • Figure 4: Membership functions used in the fuzzy evaluator. (a) Normalized sample quantity. (b) Normalized available network throughput. (c) Normalized computational capability on the local. (d) Normalized loss function training on the local dataset. (e) Evaluation mapped into different levels. In subfigure (a) - (d), the red line represents that the participant has better performance (or owns more resources) on a specific factor. The green/blue lines mean that the participant has average/poor performance (or owns average/less resource) on the specific factor. The dashed line in subfigure (a) - (d) represents the mean value of each variable.
  • Figure 5: The center of gravity (COG).
  • ...and 4 more figures