Cross-task Calibration for Asynchronous Federated Continual Learning

Federated Continual Learning (FCL) aims to empower distributed devices to learn a sequence of tasks over time. However, existing FCL research largely relies on the impractical assumption of synchronous new task arrival. This overlooks the reality of asynchronous user behavior and system latencies, forcing more efficient clients to endure costly inactivity. The practical necessity of an asynchronous method gives rise to Asynchronous Federated Continual Learning (AFCL). The server constantly receives a mixture of updates from clients at different time steps, leading to a catastrophic task drift that corrupts the global model and prevents effective learning. In this paper, we introduce a novel Cross-task Calibration framework called C$^2$-AFCL that is the first to tackle task drift at a semantic level within an Asynchronous FCL setting. Its core is a two-stage orthogonal calibration mechanism. First, intra-client calibration uses task-aware caches to mitigate variance from local client drift. Second, and more critically, inter-task interference calibration dynamically estimates an interference subspace from historical task knowledge. New updates are orthogonally projected to isolate and remove components that conflict with this subspace, preserving previous knowledge while learning new tasks. Extensive experiments show that C$^2$-AFCL significantly outperforms existing methods, demonstrating robust and efficient learning in dynamic federated environments.

Many AI systems learn from data spread across phones, sensors, or other devices, instead of collecting all data in one place. In real life, these devices do not learn new things at the same time: some users move on to new tasks earlier, while others are delayed by slower hardware or network problems. Existing methods often assume that everyone progresses together, which can leave faster devices waiting and waste useful learning time. We study what happens when devices are allowed to learn at their own pace. The main challenge is that the central model receives mixed updates from devices working on different tasks, which can pull the model in conflicting directions and make it forget what it learned before. We propose C$^2$-AFCL, a method that helps the system recognize and reduce these conflicts. Each device first uses a small memory of past tasks to make its own updates more stable. Then the server compares new updates with knowledge from earlier tasks and filters out parts that are likely to interfere. This allows the system to keep learning new tasks while better preserving old knowledge, making federated learning more practical for real-world, uneven user behavior.