Research map / Continual learning / Class-incremental
Geometry-Anchored Transport Framework for Exemplar-Free Class-Incremental Learning
Hongye Xu, Bartosz Krawczyk
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 topology-aware evolution (adapter learned in primary training) (ablation) | GATF · CIFAR-100 (from scratch, T=10) | A_last: 46.8 -> 50.3 | Improved |
| add Analytic Geometric Anchor on top of topology-aware (ablation) | GATF · CIFAR-100 (from scratch, T=10) | A_last: 50.3 -> 51.7 | Improved |
| remove post-hoc refinement (end-to-end only) (ablation) | GATF · CIFAR-100 (from scratch, T=10) | A_last: 51.7 -> 52.0 | Improved |
| remove post-hoc refinement (ablation) | GATF · TinyImageNet T=20 | A_last: 31.7 -> 31.5 | Hurt |
| add AGA to topology-aware model (ablation) | GATF · TinyImageNet T=10 | A_last: 35.6 -> 37.0 | Improved |
| use GATF instead of DPCR | GATF vs DPCR · CIFAR-100 T=10 | A_last: 50.2 -> 52.0 | Improved |
| use GATF instead of AdaGauss | GATF vs AdaGauss · ImageNet-100 T=20 | A_last: 42.6 -> 44.7 | Improved |
| use GATF instead of LDC | GATF vs LDC · ImageNet-100 T=10 | A_inc: 69.4 -> 67.4 | Hurt |
| use GATF with pretrained backbone | GATF vs EFC/AdaGauss · CUB-200 | A_last: T=10: 52.9 (AdaGauss) vs 52.4; T=20: 46.1 (EFC) vs 43.7 | No clear effect |
Closest papers by meaning
- Elastic Feature Consolidation For Cold Start Exemplar-Free Incremental Learning (ICLR 2024 · similarity 0.86)
- Reducing Class-wise Confusion for Incremental Learning with Disentangled Manifolds (CVPR 2025 · similarity 0.85)
- Geometry-Aware Optimal Transport: Fast Intrinsic Dimension and Wasserstein Distance Estimation (NeurIPS 2026 · similarity 0.85)
- Task-recency bias strikes back: Adapting covariances in Exemplar-Free Class Incremental Learning (NeurIPS 2024 · similarity 0.85)
- Geometrically Aligned Transfer Encoder for Inductive Transfer in Regression Tasks (ICLR 2024 · similarity 0.84)
- Data-Free Reservoir Features for Efficient Long-Horizon Cold-Start Continual Learning (NeurIPS 2026 · 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.
