Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Federated Learning with Confidence-Weighted Filtering and GAN-Based Completion under Noisy and Incomplete Data

Alpaslan Gokcen, Ali Boyaci

TL;DR

This work tackles federated learning under realistic data quality challenges, including noisy labels, missing classes, and non-IID client data. It presents a three-stage framework: (1) local noise cleaning using multi-metric confidence, (2) federated training of lightweight conditional GANs for class-conditioned data, and (3) data completion with synthetic samples followed by robust FL training via FedAvg or FedProx. Empirical results on MNIST and Fashion-MNIST show clear macro-F1 gains, with CleanAvg and CleanProx delivering strong baselines and GenClean variants offering additional benefits under higher noise or severe class missingness, particularly on Fashion-MNIST. The method balances privacy, practicality for edge devices, and performance, illustrating a scalable approach to robust FL in real-world, decentralized environments.

Abstract

Federated learning (FL) presents an effective solution for collaborative model training while maintaining data privacy across decentralized client datasets. However, data quality issues such as noisy labels, missing classes, and imbalanced distributions significantly challenge its effectiveness. This study proposes a federated learning methodology that systematically addresses data quality issues, including noise, class imbalance, and missing labels. The proposed approach systematically enhances data integrity through adaptive noise cleaning, collaborative conditional GAN-based synthetic data generation, and robust federated model training. Experimental evaluations conducted on benchmark datasets (MNIST and Fashion-MNIST) demonstrate significant improvements in federated model performance, particularly macro-F1 Score, under varying noise and class imbalance conditions. Additionally, the proposed framework carefully balances computational feasibility and substantial performance gains, ensuring practicality for resource constrained edge devices while rigorously maintaining data privacy. Our results indicate that this method effectively mitigates common data quality challenges, providing a robust, scalable, and privacy compliant solution suitable for diverse real-world federated learning scenarios.

Robust Federated Learning with Confidence-Weighted Filtering and GAN-Based Completion under Noisy and Incomplete Data

TL;DR

This work tackles federated learning under realistic data quality challenges, including noisy labels, missing classes, and non-IID client data. It presents a three-stage framework: (1) local noise cleaning using multi-metric confidence, (2) federated training of lightweight conditional GANs for class-conditioned data, and (3) data completion with synthetic samples followed by robust FL training via FedAvg or FedProx. Empirical results on MNIST and Fashion-MNIST show clear macro-F1 gains, with CleanAvg and CleanProx delivering strong baselines and GenClean variants offering additional benefits under higher noise or severe class missingness, particularly on Fashion-MNIST. The method balances privacy, practicality for edge devices, and performance, illustrating a scalable approach to robust FL in real-world, decentralized environments.

Abstract

Federated learning (FL) presents an effective solution for collaborative model training while maintaining data privacy across decentralized client datasets. However, data quality issues such as noisy labels, missing classes, and imbalanced distributions significantly challenge its effectiveness. This study proposes a federated learning methodology that systematically addresses data quality issues, including noise, class imbalance, and missing labels. The proposed approach systematically enhances data integrity through adaptive noise cleaning, collaborative conditional GAN-based synthetic data generation, and robust federated model training. Experimental evaluations conducted on benchmark datasets (MNIST and Fashion-MNIST) demonstrate significant improvements in federated model performance, particularly macro-F1 Score, under varying noise and class imbalance conditions. Additionally, the proposed framework carefully balances computational feasibility and substantial performance gains, ensuring practicality for resource constrained edge devices while rigorously maintaining data privacy. Our results indicate that this method effectively mitigates common data quality challenges, providing a robust, scalable, and privacy compliant solution suitable for diverse real-world federated learning scenarios.
Paper Structure (20 sections, 10 equations, 4 figures, 2 tables, 4 algorithms)

This paper contains 20 sections, 10 equations, 4 figures, 2 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Proposed Methodology
  4. Overview
  5. Stage 1: Noise Cleaning
  6. Stage 2: Collaborative GAN Training
  7. Stage 3: Data Completion and FL Training
  8. Experimental Setup
  9. Datasets and Simulation
  10. Noise Model
  11. Baseline Methods
  12. Evaluation Metrics
  13. Results and Discussion
  14. Data Quality Improvement
  15. Model Comparison under Selected Conditions
  16. ...and 5 more sections

Figures (4)

  • Figure 1: Three-Stage Federated Learning Framework for Robust Training under Noisy and Incomplete Data
  • Figure 2: F1 Score across varying noise ratios for models trained on MNIST and FashionMNIST.
  • Figure 3: F1 Score versus number of missing classes across models and datasets.
  • Figure 4: Heatmap of average F1 scores for each model on MNIST and FashionMNIST.