Research map / Federated learning and privacy / Personalized and heterogeneous FL
SUM: Unified Geometric Surgery on Spatio-Temporal Adaptation Vectors for Federated Class Incremental Learning
Jaeik Kim, Jaeyoung Do
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 Z-score trimming to spatial surgery (ablation) | SUM · CIFAR-100 (beta=0.05) | FAA: 96.32 -> 85.87 when removed | Improved |
| add spatial surgery (ablation) | SUM · CIFAR-100 (beta=0.05) | FAA: 96.32 -> 78.93 when removed | Improved |
| add temporal surgery (ablation) | SUM · CIFAR-100 (beta=0.05) | FAA: 96.32 -> 87.67 when removed | Improved |
| add sparsification of residual (ablation) | SUM · CIFAR-100 (beta=0.05) | FAA: 96.32 -> 90.23 when removed | Improved |
| add Elect-Sign (ablation) | SUM · CIFAR-100 (beta=0.05) | FAA: 96.32 -> 94.58 when removed | Improved |
| add task-specific mask modulation (ablation) | SUM · CIFAR-100 (beta=0.05) | FAA: 96.32 -> 91.24 when removed | Improved |
| project only positive-cosine components instead of both signs (ablation) | SUM · CIFAR-100 (beta=0.05) | FAA | Hurt |
| project both positive and negative cosine interactions (ablation) | SUM · CIFAR-100 (beta=0.05) | FAA | Improved |
| replace ViT with LoRA backbone | SUM · CIFAR-100 (beta=0.05) | FAA: 96.32 -> 96.35 | Improved |
| replace ViT with ConvNeXt backbone | SUM · CIFAR-100 (beta=0.05) | FAA: 96.32 -> 93.93 | Mixed |
| use SUM instead of centralized Joint training | SUM · CIFAR-100 | FAA: 92.75 -> 97.03 | Improved |
| use SUM instead of centralized Joint training | SUM · ImageNet-A | FAA: 54.64 -> 49.90 | Hurt |
Closest papers by meaning
- Federated Class-Incremental Learning: A Hybrid Approach Using Latent Exemplars and Data-Free Techniques to Address Local and Global Forgetting (ICLR 2025 · similarity 0.88)
- HERO: A Heterogeneity-Aware Benchmark Library for Federated Continual Learning (NeurIPS 2026 · similarity 0.87)
- Handling Spatial-Temporal Data Heterogeneity for Federated Continual Learning via Tail Anchor (CVPR 2025 · similarity 0.86)
- Class-wise Balancing Data Replay for Federated Class-Incremental Learning (NeurIPS 2025 · similarity 0.86)
- C$^2$Prompt: Class-aware Client Knowledge Interaction for Federated Continual Learning (NeurIPS 2025 · similarity 0.86)
- Federated Continual Learning Goes Online: Uncertainty-Aware Memory Management for Vision Tasks and Beyond (ICLR 2025 · similarity 0.85)
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.
