pFedUL: Layer-Aware Federated Unlearning for Personalized Federated Learning

2026-06-15 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

pFedUL is a novel layer-aware federated unlearning framework designed for personalized federated learning (pFL) environments, which typically decompose models into shared global and client-specific layers. Unlike traditional federated unlearning methods built for FedAvg, pFedUL addresses the unique challenge of removing data contributions while preserving personalization for remaining clients. It employs gradient-based layer-wise contribution attribution, adaptive selective unlearning strategies for different layer types, and a lightweight recalibration protocol. Evaluated on CIFAR-10, CIFAR-100, and FEMNIST, pFedUL achieves unlearning effectiveness comparable to full retraining, maintaining 97.3% personalized accuracy. It also introduces Personalization Preservation Score (PPS) and Cross-client Fairness Index (CFI) for pFL-specific unlearning quality, outperforming six state-of-the-art FU methods in personalization preservation.

Key takeaway

For MLOps engineers or AI scientists implementing GDPR-compliant data removal in personalized federated learning, pFedUL offers a robust solution. You should consider adopting its layer-aware approach to ensure effective unlearning while preserving the high personalized accuracy of remaining client models. This framework provides a practical method to balance regulatory compliance with model utility in non-IID data settings.

Key insights

Layer-aware federated unlearning balances data removal completeness with personalization preservation in pFL.

Principles

• pFL unlearning requires differentiating shared vs. personalized layer contributions.
• Unlearning completeness and personalization preservation are in inherent tension.

Method

pFedUL uses gradient-based layer-wise attribution, adaptive selective unlearning for different layer types, and a lightweight recalibration protocol for remaining clients.

In practice

• Evaluate pFL unlearning quality using Personalization Preservation Score (PPS).
• Assess cross-client fairness in unlearning with the Cross-client Fairness Index (CFI).

Topics

• Federated Learning
• Federated Unlearning
• Personalized Federated Learning
• GDPR Compliance
• Model Personalization
• Data Privacy

Best for: Research Scientist, CTO, VP of Engineering/Data, AI Scientist, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• FedMTFI: Feature Importance Based Optimized Multi Teacher Knowledge Distillation in Heterogeneous Federated Learning Environment — FedMTFI improves heterogeneous federated learning by combining multi-teacher knowledge distillation with SHAP-based feature importance for enhanced accuracy and interpretability.
• Harmonizing Multi-Objective LLM Unlearning via Unified Domain Representation and Bidirectional Logit Distillation — Multi-objective LLM unlearning can be harmonized via unified data representation and bidirectional logit distillation.
• Fed-FBD: Federated Functional Block Diversification for Isolation, Privacy, and Surgical Unlearning — Fed-FBD offers block-level isolation, privacy, and surgical unlearning in federated learning by diversifying functional blocks.
• Conflict-Aware Federated Fine-Tuning of Large Language Models with Mixture-of-Experts — FC-MoE resolves expert conflicts in federated MoE LLM fine-tuning, improving stability and performance in non-IID settings.
• To forget is to preserve: Machine Unlearning for 3D medical image segmentation — Effective machine unlearning balances data erasure with preserving model utility for retained information.
• U-CAN: Utility-Aware Contrastive Attenuation for Efficient Unlearning in Generative Recommendation — U-CAN precisely unlearns sensitive data in GenRec LLMs by selectively attenuating LoRA adapter parameters based on activation contrasts and utility.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.