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

TMLR Paper7893 Authors

Abstract: Federated continual learning (FCL) enables collaborative model training across distributed clients on sequentially arriving tasks without revisiting past data. However, existing approaches often suffer from catastrophic forgetting, rely on replay buffers or generative models that may violate privacy constraints, or assume knowledge of task identities during inference. We propose FedProTIP (Federated Projection-based Continual Learning with Task Identity Prediction), a replay-free FCL framework that maintains shared task-specific feature subspaces across clients. Each client extracts low-rank core bases from intermediate activations using randomized singular value decomposition, capturing dominant feature directions associated with the current task. These bases are transmitted to the server and aggregated to construct global task subspaces that capture shared feature directions across clients without requiring data sharing. During training, client updates are projected onto the orthogonal complement of previously learned subspaces to reduce cross-task interference and mitigate catastrophic forgetting. The learned subspaces are also reused during inference to estimate task identity via subspace relevance, enabling task-agnostic prediction without requiring explicit task labels. Experiments on CIFAR100, ImageNet-R, and DomainNet demonstrate that FedProTIP consistently outperforms state-of-the-art federated continual learning baselines while maintaining lower training time, memory footprint, and communication cost.

Submission Type: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission:

We have marked all revised parts in blue. The changes made since the initial draft are summarized as follows:
- Corrected FOT description in Sections 2.2 and 4 (randomized sketches under secure aggregation).
- Replaced Appendix F.2 with a SecAgg-compatible weighted Gaussian sketch (Table 19, Table 20).
- Added Figure 2 with empirical estimates of the projection-dependent quantities and revised Section 5 Remarks.
- Rewrote Equation (8) in Section 4.2 with explicit relative-energy denominator.
- Added Figure 4(a) and Table 14 with per-task TIP accuracy; expanded Section 6.1 discussion.
- Added paragraph at end of Section 4.4 on TIP's distinct-distribution assumption.
- Expanded Section 6.2 with DomainNet shared-head versus task-specific-head trade-off.
- Updated Table 1 caption and Section 6.1 to clarify task-agnostic versus task-aware reporting.
- Restructured Section 6.5 into asymptotic complexity (Table 4) and empirical measurements.
- Added parameter-efficient FCL discussion in Appendix F.1.

Assigned Action Editor: ~Liyuan_Wang1

Submission Number: 7893