FedXDS: Leveraging Model Attribution Methods to counteract Data Heterogeneity in Federated Learning

2026-06-30 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, quick

Summary

FedXDS (Federated Learning via XAI-guided Data Sharing) is a novel approach that leverages Explainable AI (XAI) methods to address data heterogeneity in federated learning. While federated learning allows collaborative model training without raw data sharing, performance often degrades with statistically heterogeneous client data. FedXDS is the first method to use feature attribution techniques, specifically propagation-based attribution, to identify and selectively share task-relevant data elements between clients. This process, achieved through a single backward pass, aims to mitigate heterogeneity and align client contributions. To protect sensitive information, FedXDS integrates metric privacy techniques, offering formal privacy guarantees while maintaining utility. Experimental results show FedXDS consistently achieves higher accuracy and faster convergence than existing methods across various client numbers and heterogeneity settings, demonstrating robustness against membership inference and feature inversion attacks.

Key takeaway

For Machine Learning Engineers developing federated learning systems, FedXDS offers a robust solution to data heterogeneity challenges. You should consider integrating XAI-guided data sharing to improve model accuracy and convergence speed, especially with diverse client data. Implementing metric privacy ensures formal privacy guarantees for shared data, enhancing security against inference and inversion attacks. This approach allows you to maintain utility while protecting sensitive information.

Key insights

FedXDS uses XAI feature attribution to selectively share data, mitigating heterogeneity in federated learning while preserving privacy.

Principles

• XAI can extend beyond transparency to solve ML challenges.
• Selective data sharing improves federated learning performance.
• Metric privacy offers formal guarantees for shared data.

Method

FedXDS employs propagation-based attribution via a single backward pass to identify task-relevant features, enabling selective data sharing between clients to reduce data heterogeneity.

In practice

• Implement XAI for data selection in FL.
• Apply metric privacy to shared data.
• Evaluate robustness against inference attacks.

Topics

• Federated Learning
• Explainable AI
• Data Heterogeneity
• Model Attribution
• Metric Privacy
• Data Sharing

Code references

• MaxH1996/FedXDS

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Security Engineer

Related on AIssential

• A Survey on Federated Causal Discovery and Inference — Federated Causal Discovery and Inference are complementary stages in a unified pipeline for privacy-preserving causal reasoning.
• HASA: Subnet Allocation for Compute-Constrained Model-Heterogeneous Federated Learning — HASA improves federated learning performance by dynamically allocating subnet widths based on client data heterogeneity.
• Federated Gaussian Process Learning via Pseudo-Representations for Large-Scale Multi-Robot Systems — pxpGP enables scalable, privacy-preserving Gaussian Process learning for large multi-robot systems via pseudo-representations and optimized ADMM.
• FedIDM: Achieving Fast and Stable Convergence in Byzantine Federated Learning through Iterative Distribution Matching — FedIDM uses distribution matching and condensed data to achieve fast, stable, and robust federated learning convergence.
• Enabling privacy-preserving AI training on everyday devices — FTTE accelerates federated learning on resource-constrained edge devices by optimizing memory, communication, and update synchronization.
• Statistical Limits and Efficient Algorithms for Differentially Private Federated Learning — New methods improve federated learning accuracy and communication efficiency while preserving differential privacy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.