Pattern Recognition Tasks with Personalized Federated Learning

This study explores personalized federated learning (PFL) as a solution for pattern recognition tasks across decentralized data sources. Traditional federated learning (FL) approaches face challenges in handling non-IID (non-independent and identically distributed) data, which often reduces performance in tasks requiring personalization. To address this, we propose a PFL framework that leverages local adaptation and global collaboration to enhance model accuracy while preserving data privacy.The proposed method is evaluated on image classification, speech recognition, and text analysis tasks, demonstrating improved performance compared to centralized and standard FL approaches. By incorporating personalized model updates, the system adapts to client-specific patterns while benefiting from shared knowledge, ensuring robustness and generalizability across diverse datasets. This research highlights the potential of PFL in enabling secure, scalable, and personalized solutions for pattern recognition applications.

In PLOS ONE