Research map / Continual learning / Online and replay-based
Task-Aware Prompt Gradient Projection for Parameter-Efficient Tuning Federated Class-Incremental Learning
Hualong Ke, Jiangming Shi, Yachao Zhang, Fangyong Wang, Yuan Xie, Yanyun Qu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add TAPGP gradient projection (ablation) | LANDER+VPT-Deep · CIFAR-100 (NIID(0.5), T=5) | average accuracy: 64.58 -> 69.11 | Improved |
| add inter-task prompt replay (ablation) | LANDER+VPT-Deep · CIFAR-100 (NIID(0.5), T=5) | average accuracy: 64.58 -> 67.72 | Improved |
| add intra-task prompt replay (ablation) | LANDER+VPT-Deep · CIFAR-100 (NIID(0.5), T=5) | average accuracy: 64.58 -> 66.95 | Improved |
| add both intra- and inter-task replay (DLPR) (ablation) | LANDER+VPT-Deep · CIFAR-100 (NIID(0.5), T=5) | average accuracy: 64.58 -> 70.45 | Improved |
| add DLPR on top of TAPGP (ablation) | TPPR · CIFAR-100 (NIID(0.5), T=5) | average accuracy: 69.11 -> 73.84 | Improved |
| add TAPGP on top of DLPR (ablation) | TPPR · CIFAR-100 (NIID(0.5), T=5) | average accuracy: 70.45 -> 73.84 | Improved |
| add inter-task replay to TAPGP+intra (ablation) | TPPR · CIFAR-100 (NIID(0.5), T=5) | forgetting F: 16.83 -> 11.30 | Improved |
| replace LANDER with VPT-Deep prompt tuning backbone (ablation) | LANDER · CIFAR-100 (NIID(0.5), T=5) | average accuracy: 48.23 -> 64.58 | Improved |
| add VPT-Deep to LANDER (ablation) | LANDER · CIFAR-100 (IID, T=5) | forgetting F: 18.03 -> 27.74 | Hurt |
| add TPPR modules to LANDER+VPT-Deep (ablation) | LANDER+VPT-Deep · Tiny-ImageNet (IID, T=10) | average accuracy: 42.74 -> 65.96 | Improved |
| start gradient projection at deeper layer (us=6 vs us=0) (ablation) | TAPGP · CIFAR-100 (NIID(0.5), T=5) | average accuracy | Mixed |
| mixing prompts instead of calculating attention for inter-task replay (ablation) | DLPR · CIFAR-100 (NIID(0.5)) | average accuracy | Improved |
Closest papers by meaning
- Federated Continual Learning via Prompt-based Dual Knowledge Transfer (ICML 2024 · similarity 0.90)
- C$^2$Prompt: Class-aware Client Knowledge Interaction for Federated Continual Learning (NeurIPS 2025 · similarity 0.89)
- CAN: Leveraging Clients As Navigators for Generative Replay in Federated Continual Learning (ICML 2025 · similarity 0.88)
- Class-wise Balancing Data Replay for Federated Class-Incremental Learning (NeurIPS 2025 · similarity 0.88)
- Accurate Forgetting for Heterogeneous Federated Continual Learning (ICLR 2024 · similarity 0.87)
- Global Prompt Refinement with Non-Interfering Attention Masking for One-Shot Federated Learning (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
