Research map / Continual learning / Online and replay-based
Less Is More in Federated Continual Learning: RieSelect for Conflict-Aware Layer Selection in LLMs
Wenqi Qiu, Yipeng Zhou, Lin Zhu, Laizhong Cui
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 Step 2 Riemannian safe-step adjustment (ablation) | RieSelect · SuperNI Order 2 (LLaMA2-7B) | AA: 32.98 -> 44.46 | Improved |
| add Step 2 Riemannian safe-step adjustment (ablation) | RieSelect · SuperNI Order 2 (LLaMA2-7B) | BWT: -15.44 -> 5.08 | Improved |
| add Step 3 knapsack layer selection (ablation) | RieSelect · SuperNI Order 2 (LLaMA2-7B) | AA: 34.46 -> 44.46 | Improved |
| add Step 3 knapsack layer selection (ablation) | RieSelect · SuperNI Order 2 (LLaMA2-7B) | BWT: 2.32 -> 5.08 | Improved |
| add Step 2 Riemannian safe-step adjustment (ablation) | RieSelect · LongSeq Order 3 (LLaMA2-7B) | AA: 35.23 -> 73.39 | Improved |
| add Step 2 Riemannian safe-step adjustment (ablation) | RieSelect · LongSeq Order 3 (LLaMA2-7B) | BWT: -38.18 -> 6.13 | Improved |
| add Step 3 knapsack layer selection (ablation) | RieSelect · LongSeq Order 3 (LLaMA2-7B) | AA: 39.96 -> 73.39 | Improved |
| add Step 3 knapsack layer selection (ablation) | RieSelect · LongSeq Order 3 (LLaMA2-7B) | BWT: 3.79 -> 6.13 | Improved |
| set safe-basin radius R_eps to 10 instead of smaller or larger values (ablation) | RieSelect · SuperNI Order 2 (LLaMA2-7B) | AA and BWT: R_eps = 10 best AA and only positive BWT | Mixed |
| upload all layers instead of a budget-matched Top-K subset (increase transmitted-layer ratio beyond moderate level) | LoRA-based FCL · Transmitted-layer ratio sweep | AA | No clear effect |
| revert the top-conflict uploaded layers versus random layers (ablation) | Full-uplink LoRA FCL · Reversion study (Fig. 1c-d) | BWT and new-task accuracy | Improved |
| apply budget-matched Top-K layer upload to baselines | baselines (LoRA, EWC, A-GEM, etc.) · LLaMA2-7B Order 3 matched budget | Final AA | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- C-LoRA — C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- HERO: A Heterogeneity-Aware Benchmark Library for Federated Continual Learning (NeurIPS 2026 · similarity 0.86)
- Accurate Forgetting for Heterogeneous Federated Continual Learning (ICLR 2024 · similarity 0.85)
- Resource-Constrained Federated Continual Learning: What Does Matter? (NeurIPS 2025 · similarity 0.85)
- Federated Continual Learning Goes Online: Uncertainty-Aware Memory Management for Vision Tasks and Beyond (ICLR 2025 · similarity 0.85)
- C$^2$Prompt: Class-aware Client Knowledge Interaction for Federated Continual Learning (NeurIPS 2025 · similarity 0.84)
- Rethinking the Stability-Plasticity Trade-off in Continual Learning from an Architectural Perspective (ICML 2025 · similarity 0.84)
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.
