Research map / Continual learning / Class-incremental
Topology-Preserving Class-Incremental Learning
Xiaoyu Tao, Xinyuan Chang, Xiaopeng Hong, Xing Wei, Yihong Gong
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 |
|---|---|---|---|
| use TPCIL instead of LUCIR | class-incremental CNN · CIFAR100 (5 sessions) | average accuracy (%): 63.42 -> 65.34 | Improved |
| use TPCIL instead of LUCIR | class-incremental CNN · subImageNet (10 sessions) | average accuracy (%): 68.09 -> 74.81 | Improved |
| use TPCIL instead of LUCIR | class-incremental CNN · ImageNet (5 sessions) | average accuracy (%): 64.34 -> 64.89 | Improved |
| use TPL instead of feature distillation loss (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 60.55 -> 65.34 | Improved |
| use TPL instead of logit distillation loss (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 55.61 -> 65.34 | Improved |
| use correlation-form TPL instead of l1 form (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 64.17 -> 65.34 | Improved |
| use correlation-form TPL instead of l2 form (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 62.75 -> 65.34 | Improved |
| add logit distillation loss to TPL (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 65.34 -> 56.19 | Hurt |
| add feature distillation loss to TPL (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 65.34 -> 58.66 | Hurt |
| use EHG vertices instead of random exemplars (1 per class) (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 33.86 -> 42.26 | Improved |
| use EHG vertices instead of random exemplars (20 per class) (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 64.70 -> 65.34 | Improved |
| add TPL term (lambda 0 -> 15) (ablation) | TPCIL · CIFAR100 (5 sessions) | average accuracy (%): 22.34 -> 65.34 | Improved |
Closest papers by meaning
- Topology-aware Knowledge Preservation for Class-Incremental Learning (AAAI 2026 · similarity 0.90)
- What Matters in Graph Class Incremental Learning? An Information Preservation Perspective (NeurIPS 2024 · similarity 0.87)
- Reducing Class-wise Confusion for Incremental Learning with Disentangled Manifolds (CVPR 2025 · similarity 0.86)
- Order-Robust Class Incremental Learning: Graph-Driven Dynamic Similarity Grouping (CVPR 2025 · similarity 0.85)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.85)
- DYSON: Dynamic Feature Space Self-Organization for Online Task-Free Class Incremental Learning (CVPR 2024 · 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.
