Task-Agnostic Federated Continual Learning via Replay-Free Gradient Projection

2025-09-25

AI Summary

Federated continual learning (FCL) confronts two intertwined challenges: exacerbated catastrophic forgetting under data heterogeneity, communication constraints, and privacy requirements; and inference difficulty due to unknown task identities. To address these, we propose a replay-free gradient orthogonal projection framework. On each client, model update gradients are orthogonally projected onto the complement of the historical task representation subspace—thereby mitigating cross-task interference and preserving prior knowledge. Concurrently, we introduce a lightweight core-basis-driven task identity prediction module that enables adaptive inference without prior task identifiers. The approach ensures low communication overhead and strong privacy preservation. Evaluated on multiple standard FCL benchmarks, our method consistently outperforms existing state-of-the-art approaches, achieving significant average accuracy gains—particularly in the challenging task-identity-unknown setting.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Natural Language Processing: Learning & Optimization for NLP
• Search and Optimization: Learning to Search

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
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Federated continual learning (FCL) enables distributed client devices to learn from streaming data across diverse and evolving tasks. A major challenge to continual learning, catastrophic forgetting, is exacerbated in decentralized settings by the data heterogeneity, constrained communication and privacy concerns. We propose Federated gradient Projection-based Continual Learning with Task Identity Prediction (FedProTIP), a novel FCL framework that mitigates forgetting by projecting client updates onto the orthogonal complement of the subspace spanned by previously learned representations of the global model. This projection reduces interference with earlier tasks and preserves performance across the task sequence. To further address the challenge of task-agnostic inference, we incorporate a lightweight mechanism that leverages core bases from prior tasks to predict task identity and dynamically adjust the global model's outputs. Extensive experiments across standard FCL benchmarks demonstrate that FedProTIP significantly outperforms state-of-the-art methods in average accuracy, particularly in settings where task identities are a priori unknown.

Problem

• Mitigates catastrophic forgetting in federated continual learning systems
• Addresses data heterogeneity and communication constraints in decentralized settings
• Enables task-agnostic inference without requiring prior task identities

Innovation

• Uses gradient projection to mitigate forgetting
• Projects updates onto orthogonal complement space
• Predicts task identity with lightweight core bases