Research map / Continual learning / Class-incremental
Bring Evanescent Representations to Life in Lifelong Class Incremental Learning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| jointly model feature and semantic drift and fuse their estimates | CIFAR100 class-incremental learning · CIFAR100, 5, 10, and 20 steps | per-class average top-1 accuracy: 1.75%, 4.6%, and 3.07% | Helped |
| jointly model feature and semantic drift and fuse their estimates | TinyImageNet class-incremental learning · TinyImageNet, 5, 10, and 20 steps | per-class average top-1 accuracy: 48.74 vs 47.89; 46.92 vs 45.41; 44.61 vs 41.46 | Helped |
| jointly model feature and semantic drift and fuse their estimates | CUB200 class-incremental learning · CUB200, 5, 10, and 20 steps | per-class average top-1 accuracy: 56.97 vs 52.30; 52.58 vs 38.30; 38.26 vs 18.17 | Helped |
| add joint fusion of feature and semantic drift instead of using either drift alone (ablation) | CIFAR100 class-incremental learning · CIFAR100, 5 steps | per-class average top-1 accuracy: 57.91 -> 59.37; 58.33 -> 59.37 | Helped |
| add joint fusion of feature and semantic drift instead of using either drift alone (ablation) | CUB200 class-incremental learning · CUB200, 20 steps | per-class average top-1 accuracy: 31.36 -> 37.99; 32.50 -> 38.26 | Helped |
| add the correlation objective to the fusion loss (ablation) | GM-MLP drift fusion · CIFAR100, 5 steps | per-class average top-1 accuracy: 58.89 -> 59.37 | Helped |
| add the correlation objective to the fusion loss (ablation) | VM-VAE drift fusion · CUB200, 20 steps | per-class average top-1 accuracy: 36.80 -> 38.26 | Helped |
| replace Gaussian MLP drift models with variational VAE drift models (ablation) | fused drift estimation · CIFAR100 and CUB200 class-incremental setups | per-class average top-1 accuracy: CIFAR100, 5 steps: 59.37 vs 58.72; CUB200, 20 steps: 37.99 vs 38.26 | Mixed |
| use prototype drift compensation as in SDC | CUB200 class-incremental learning · CUB200, 5, 10, and 20 steps | per-class average top-1 accuracy | No clear effect |
Closest papers by meaning
- Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning (ICML 2025 · similarity 0.90)
- Long-Tail Class Incremental Learning via Independent Sub-prototype Construction (CVPR 2024 · similarity 0.89)
- Prospective Representation Learning for Non-Exemplar Class-Incremental Learning (NeurIPS 2024 · similarity 0.89)
- eTag: Class-Incremental Learning via Embedding Distillation and Task-Oriented Generation (AAAI 2024 · similarity 0.89)
- Adaptive Discovering and Merging for Incremental Novel Class Discovery (AAAI 2024 · similarity 0.88)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.88)
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.