Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cost-TrustFL: Cost-Aware Hierarchical Federated Learning with Lightweight Reputation Evaluation across Multi-Cloud

Jixiao Yang, Jinyu Chen, Zixiao Huang, Chengda Xu, Chi Zhang, Sijia Li

TL;DR

Cost-TrustFL tackles the triad of non-IID data, Byzantine threats, and costly cross-cloud communication in multi-cloud federated learning. It introduces a hierarchical aggregation framework, a gradient-based Shapley approximation with linear complexity, and a cost-aware client selection mechanism to jointly optimize accuracy and communication cost. Key contributions include a lightweight reputation scheme, intra-cloud prioritization to reduce egress, and FLTrust-inspired robustness with reference gradients. The approach yields robust accuracy improvements (e.g., 86.7% under aggressive attacks) and significant cross-cloud cost reductions (about 32%), demonstrating practical viability for real-world multi-cloud FL deployments.

Abstract

Federated learning across multi-cloud environments faces critical challenges, including non-IID data distributions, malicious participant detection, and substantial cross-cloud communication costs (egress fees). Existing Byzantine-robust methods focus primarily on model accuracy while overlooking the economic implications of data transfer across cloud providers. This paper presents Cost-TrustFL, a hierarchical federated learning framework that jointly optimizes model performance and communication costs while providing robust defense against poisoning attacks. We propose a gradient-based approximate Shapley value computation method that reduces the complexity from exponential to linear, enabling lightweight reputation evaluation. Our cost-aware aggregation strategy prioritizes intra-cloud communication to minimize expensive cross-cloud data transfers. Experiments on CIFAR-10 and FEMNIST datasets demonstrate that Cost-TrustFL achieves 86.7% accuracy under 30% malicious clients while reducing communication costs by 32% compared to baseline methods. The framework maintains stable performance across varying non-IID degrees and attack intensities, making it practical for real-world multi-cloud deployments.

Cost-TrustFL: Cost-Aware Hierarchical Federated Learning with Lightweight Reputation Evaluation across Multi-Cloud

TL;DR

Cost-TrustFL tackles the triad of non-IID data, Byzantine threats, and costly cross-cloud communication in multi-cloud federated learning. It introduces a hierarchical aggregation framework, a gradient-based Shapley approximation with linear complexity, and a cost-aware client selection mechanism to jointly optimize accuracy and communication cost. Key contributions include a lightweight reputation scheme, intra-cloud prioritization to reduce egress, and FLTrust-inspired robustness with reference gradients. The approach yields robust accuracy improvements (e.g., 86.7% under aggressive attacks) and significant cross-cloud cost reductions (about 32%), demonstrating practical viability for real-world multi-cloud FL deployments.

Abstract

Federated learning across multi-cloud environments faces critical challenges, including non-IID data distributions, malicious participant detection, and substantial cross-cloud communication costs (egress fees). Existing Byzantine-robust methods focus primarily on model accuracy while overlooking the economic implications of data transfer across cloud providers. This paper presents Cost-TrustFL, a hierarchical federated learning framework that jointly optimizes model performance and communication costs while providing robust defense against poisoning attacks. We propose a gradient-based approximate Shapley value computation method that reduces the complexity from exponential to linear, enabling lightweight reputation evaluation. Our cost-aware aggregation strategy prioritizes intra-cloud communication to minimize expensive cross-cloud data transfers. Experiments on CIFAR-10 and FEMNIST datasets demonstrate that Cost-TrustFL achieves 86.7% accuracy under 30% malicious clients while reducing communication costs by 32% compared to baseline methods. The framework maintains stable performance across varying non-IID degrees and attack intensities, making it practical for real-world multi-cloud deployments.
Paper Structure (25 sections, 13 equations, 7 figures, 2 tables, 1 algorithm)

This paper contains 25 sections, 13 equations, 7 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Byzantine-Robust Federated Learning
  4. Contribution Evaluation in FL
  5. Communication-Efficient FL
  6. Problem Formulation
  7. System Model
  8. Threat Model
  9. Cost Model
  10. Optimization Objective
  11. Methodology
  12. Hierarchical Aggregation Architecture
  13. Gradient-Based Shapley Approximation
  14. Cost-Aware Client Selection
  15. Byzantine-Robust Aggregation
  16. ...and 10 more sections

Figures (7)

  • Figure 1: Cost-TrustFL architecture showing hierarchical aggregation across multi-cloud with edge aggregators minimizing cross-cloud communication. Malicious client (red) is identified through reputation evaluation.
  • Figure 2: Convergence comparison under 30% malicious clients with label flipping attacks on (a) CIFAR-10 and (b) FEMNIST datasets.
  • Figure 3: (a) Cost-accuracy trade-off showing Pareto improvement of Cost-TrustFL. (b) Cost breakdown by component.
  • Figure 4: (a) Performance under varying malicious client ratios. (b) Sensitivity to non-IID degree (Dirichlet $\alpha$).
  • Figure 5: (a) Computation time comparison for different Shapley methods. (b) Correlation between gradient-based estimation and true Shapley values.
  • ...and 2 more figures