Research map / Continual learning / Class-incremental
Beyond Point-wise Neural Collapse: A Topology-Aware Hierarchical Classifier for Class-Incremental Learning
HuiYu Yi, Xu Zhiming, Dunwei Tu, Zhicheng Wang, Baile Xu, Furao Shen
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 |
|---|---|---|---|
| replace original classifier with HC-SOINN in seven PTM-based CIL methods | SEMA · Split CUB-200 (20 tasks) | last-task accuracy (%): 65.56 -> 86.56 | Improved |
| replace FC classifier with HC-SOINN | CODA-Prompt · Split CUB-200 (20 tasks) | last-task accuracy (%): 69.97 -> 85.75 | Improved |
| replace classifier with HC-SOINN | CL-LoRA · Split CUB-200 (20 tasks) | last-task accuracy (%): 74.47 -> 74.60 | No clear effect |
| replace classifier with HC-SOINN | EASE · Split CUB-200 (20 tasks) | last-task accuracy (%): 81.21 -> 81.13 | No clear effect |
| replace classifier with HC-SOINN | CODA-Prompt · Split CIFAR-100 (10 tasks) | last-task accuracy (%): 86.96 -> 87.56 | Improved |
| replace classifier with HC-SOINN | DualPrompt · Split ImageNet-R (40 tasks) | last-task accuracy (%): 58.67 -> 69.27 | Improved |
| add STAR drift alignment to HC-SOINN (ablation) | DualPrompt + HC-SOINN · Split CUB-200 (20 tasks) | last-task accuracy (%): 77.31 -> 85.84 | Improved |
| add STAR drift alignment to HC-SOINN (ablation) | DualPrompt + HC-SOINN · Split ImageNet-R (40 tasks) | last-task accuracy (%): 69.27 -> 69.75 | No clear effect |
| add STAR drift alignment to HC-SOINN (ablation) | CODA-Prompt + HC-SOINN · Split CIFAR-100 (10 tasks) | last-task accuracy (%): 87.56 -> 89.67 | Improved |
| use HC-SOINN instead of KAC classifier | DualPrompt · Split ImageNet-R (40 tasks) | last-task accuracy (%): 63.85 -> 69.75 (with STAR) | Improved |
| KAC classifier instead of FC classifier | DualPrompt · Split CIFAR-100 (10 tasks) | average accuracy (%): 86.40 -> 86.30 | Hurt |
| combine hierarchical clustering with SOINN refinement versus either alone (ablation) | HC-SOINN · Split CUB-200 (CODA-Prompt) | last-task accuracy (%): 82.65 / 83.80 -> 85.75 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CL-LoRA — CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning (CVPR 2025) · better on 3 of 3 shared results
- DualPrompt — DualPrompt: Complementary Prompting for Rehearsal-Free Continual Learning (ECCV 2022) · 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
- DYSON: Dynamic Feature Space Self-Organization for Online Task-Free Class Incremental Learning (CVPR 2024 · similarity 0.87)
- Guiding Neural Collapse: Optimising Towards the Nearest Simplex Equiangular Tight Frame (NeurIPS 2024 · similarity 0.87)
- Neural Collapse Beyond the Unconstrained Features Model: Landscape, Dynamics, and Generalization in the Mean-Field Regime (ICML 2025 · similarity 0.87)
- Neural Collapse in Cumulative Link Models for Ordinal Regression: An Analysis with Unconstrained Feature Model (NeurIPS 2025 · similarity 0.87)
- Reducing Class-wise Confusion for Incremental Learning with Disentangled Manifolds (CVPR 2025 · similarity 0.86)
- Compress to One Point: Neural Collapse for Pre-Trained Model-Based Class-Incremental Learning (AAAI 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.
