Lazy But Effective: Collaborative Personalized Federated Learning with Heterogeneous Data

Ljubomir Rokvic, Panayiotis Danassis, Boi Faltings·May 05, 2025

Summary

PFedLIA, a personalized federated learning framework, tackles heterogeneity in client data distributions by clustering clients using 'Lazy Influence' for efficient model aggregation. This enables collaborative global model training that captures specific client patterns, outperforming existing baselines, including a 17% improvement on CIFAR100. The method involves client-centric personalized learning, allowing selection of relevant peers for model aggregation. The summary highlights 51 key references in mobile edge networks, federated learning, and personalized models, covering topics like efficient kernel methods, multi-center federated learning, and privacy-preserving data quality evaluation.

Introduction

Background

Overview of federated learning

Challenges in federated learning: heterogeneity in client data distributions

Importance of personalized federated learning

Objective

To introduce PFedLIA, a novel personalized federated learning framework

To address the issue of heterogeneity in client data distributions through clustering

To demonstrate the framework's ability to capture specific client patterns and improve model performance

Method

Client Clustering

Explanation of 'Lazy Influence' technique for clustering clients

How clustering enables efficient model aggregation by grouping similar clients

Personalized Learning

Client-centric personalized learning approach

Selection of relevant peers for model aggregation based on client-specific patterns

Model Aggregation

Process of aggregating models from clustered clients

How aggregation captures and leverages client-specific patterns for improved performance

Results

Performance Evaluation

Comparison with existing baselines, including a 17% improvement on CIFAR100

Detailed metrics and benchmarks used for evaluation

Case Studies

Illustrative examples demonstrating the framework's effectiveness in real-world scenarios

References

Mobile Edge Networks

Efficient kernel methods for federated learning

Multi-center federated learning strategies

Federated Learning

Techniques for handling heterogeneity in client data

Optimization methods for federated learning models

Personalized Models

Privacy-preserving data quality evaluation methods

Personalization techniques in federated learning

Additional Topics

Literature on mobile edge computing and its integration with federated learning

Research on improving model performance through client-specific adaptations

Basic info

papers

machine learning

artificial intelligence

Advanced features