Research map / Continual learning / Class-incremental
Multi-Anchor Distillation with Text-Guided Analytic Classifier for Continual Learning
Qier Meng, De Cheng, Jiahao Li, Cheng Deng
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 multi-anchor distillation (ablation) | analytic classifier + CMR · CIFAR-100 | Last accuracy: 81.06 -> 83.70 | Improved |
| add multi-anchor distillation (ablation) | analytic classifier + CMR · ImageNet-R | Last accuracy: 82.90 -> 84.69 | Improved |
| add analytic classifier on top of multi-anchor distillation (ablation) | MAD · CIFAR-100 | Last accuracy: 79.95 -> 80.97 | Improved |
| add cross-modal regularization (text prior + subspace) (ablation) | MAD + analytic classifier · CIFAR-100 | Last accuracy: 80.97 -> 83.70 | Improved |
| add cross-modal regularization (text prior + subspace) (ablation) | MAD + analytic classifier · CUB-200 | Last accuracy: 76.13 -> 78.63 | Improved |
| add multi-anchor distillation only (ablation) | no components · CIFAR-100 | Average accuracy: 66.68 -> 88.63 | Improved |
| add text prior (ablation) | analytic classifier · CIFAR-100 | Last accuracy: 80.42 -> 82.15 | Improved |
| add text subspace regularization (ablation) | analytic classifier · CIFAR-100 | Last accuracy: 80.42 -> 81.28 | Improved |
| add both text prior and subspace regularization (ablation) | analytic classifier · CIFAR-100 | Last accuracy: 80.42 -> 83.70 | Improved |
| multi-anchor instead of single-anchor distillation (ablation) | distillation · CIFAR-100 (10 tasks) | cumulative feature drift: 0.016 -> 0.006 | Improved |
| add distillation (ablation) | encoder · CIFAR-100 (10 tasks) | cumulative feature drift: 0.050 -> 0.006 | Improved |
| increase number of anchors beyond M=5 (ablation) | multi-anchor distillation · CIFAR-100 | Last accuracy | No clear effect |
Closest papers by meaning
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.88)
- Synthetic Data is an Elegant GIFT for Continual Vision-Language Models (CVPR 2025 · similarity 0.86)
- Improving Continual Learning Performance and Efficiency with Auxiliary Classifiers (ICML 2025 · similarity 0.86)
- Hierarchical Semantic Tree Anchoring for CLIP-Based Class-Incremental Learning (NeurIPS 2026 · similarity 0.86)
- Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- Adapt-$\infty$: Scalable Continual Multimodal Instruction Tuning via Dynamic Data Selection (ICLR 2025 · 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.
