DOLFIN: Balancing Stability and Plasticity in Federated Continual Learning

2025-10-15

AI Summary

Federated continual learning (FCL) faces the fundamental trade-off between stability and plasticity, while simultaneously struggling to ensure privacy preservation and communication efficiency. To address these challenges, we propose DOLFIN: a vision Transformer–based framework that introduces Orthogonal Low-Rank Adaptation (Orthogonal LoRA) for lightweight incremental model updates, and a Dual Gradient Projection Memory (DualGPM) mechanism that dynamically constrains local parameter update directions to retain prior knowledge. DOLFIN natively supports differential privacy and low-bandwidth communication by transmitting only adapter parameters—not full models. Evaluated on four non-i.i.d. benchmarks—including CIFAR-100 and ImageNet-R—DOLFIN consistently outperforms six state-of-the-art baselines, achieving average accuracy gains of 2.1–5.7% while maintaining memory overhead comparable to baseline methods.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Federated continual learning (FCL) enables models to learn new tasks across multiple distributed clients, protecting privacy and without forgetting previously acquired knowledge. However, current methods face challenges balancing performance, privacy preservation, and communication efficiency. We introduce a Distributed Online LoRA for Federated INcremental learning method DOLFIN, a novel approach combining Vision Transformers with low-rank adapters designed to efficiently and stably learn new tasks in federated environments. Our method leverages LoRA for minimal communication overhead and incorporates DualGradient Projection Memory (DualGPM) to prevent forgetting. Evaluated on CIFAR-100, ImageNet-R, ImageNet-A, and CUB-200 under two Dirichlet heterogeneity settings, DOLFIN consistently surpasses six strong baselines in final average accuracy while matching their memory footprint. Orthogonal low-rank adapters offer an effective and scalable solution for privacy-preserving continual learning in federated settings.

Problem

• Balancing stability and plasticity in federated continual learning systems
• Addressing privacy preservation and communication efficiency challenges
• Preventing catastrophic forgetting while learning new distributed tasks

Innovation

• Combines Vision Transformers with low-rank adapters
• Uses LoRA for minimal communication overhead
• Incorporates DualGPM to prevent catastrophic forgetting