Research map / Continual learning / Class-incremental
Dual-Estimator: Decoupling Global and Local Semantic Shift for Drift Compensation in Class-Incremental Learning
Fankang Xu, Lu Jin, Yanpeng Sun, Shiyu Xuan, Zechao Li
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 MoE-E on top of the base distillation model (ablation) | Base (Lcls + Lkd) · TinyImageNet (20 tasks) | A_inc (%): 36.40 -> 43.55 | Improved |
| add LoR-E on top of the base distillation model (ablation) | Base (Lcls + Lkd) · TinyImageNet (20 tasks) | A_inc (%): 36.40 -> 39.49 | Improved |
| add LoR-E alone on top of the base model (ablation) | Base (Lcls + Lkd) · ImageNet-Subset (20 tasks) | A_inc (%): 55.17 -> 54.17 | Hurt |
| add LoR-E to MoE-E (combine global and local estimators) (ablation) | Dual-E · TinyImageNet (20 tasks) | A_inc (%): 43.55 -> 44.36 | Improved |
| add historical preservation term in LoR-E (ablation) | Dual-E · TinyImageNet (20 tasks) | A_inc (%): 44.23 -> 44.36 | Improved |
| add rotation-based pixel fusion augmentation for MoE-E experts (ablation) | Dual-E · TinyImageNet (20 tasks) | A_last (%): 3.88 -> 30.35 | Improved |
| add HP weight alpha greater than 0 (alpha=5) (ablation) | LoR-E · CIFAR-100 / ImageNet-Subset (10 tasks) | accuracy | Improved |
| vary number of experts K and rank r (ablation) | Dual-E · CIFAR-100 (10 tasks) | accuracy | No clear effect |
| plug Dual-E into PASS instead of no compensation or LDC/DP | PASS · TinyImageNet (20 tasks) | A_inc (%): 32.01 -> 39.75 | Improved |
| plug Dual-E into NAPA-VQ | NAPA-VQ · ImageNet-Subset (20 tasks) | A_inc (%): 24.38 -> 49.61 | Improved |
| use Dual-E instead of DP on long 50-task sequences | drift compensation · ImageNet-1K (50 tasks) | A_inc (%): 38.88 -> 40.13 | Improved |
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.
- NAPA-VQ — NAPA-VQ: Neighborhood-Aware Prototype Augmentation with Vector Quantization for Continual Learning (ICCV 2023) · better on 1 of 1 shared result
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
- Don't Forget Why You Started: Tackling Dual Forgetting in Vision-Language Continual Learning (ICML 2026 · similarity 0.89)
- Semantic Shift Estimation via Dual-Projection and Classifier Reconstruction for Exemplar-Free Class-Incremental Learning (ICML 2025 · similarity 0.89)
- Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning (ICML 2025 · similarity 0.89)
- DGS: Dual Gradient and Semantic-Shift Guided Low-Rank Adaptation for Class Incremental Learning (CVPR 2026 · similarity 0.88)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.86)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.86)
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.
