Research map / Continual learning / Class-incremental
Few-Shot Class-Incremental Learning via Entropy-Regularized Data-Free Replay
Huan Liu, Li Gu, Zhixiang Chi, Yang Wang, Yuanhao Yu, Jun Chen, Jin Tang
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 entropy regularization to generator training (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | average accuracy: 60.39 -> 60.77 | Improved |
| add entropy regularization to generator training (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | final session accuracy: 49.50 -> 50.14 | Improved |
| add re-labeling of generated data with one-hot labels (instead of KL distillation) (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | average accuracy: 59.74 -> 60.77 | Improved |
| add re-labeling of generated data with one-hot labels (instead of KL distillation) (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | final session accuracy: 48.92 -> 50.14 | Improved |
| add backbone fine-tuning during incremental sessions (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | average accuracy: 59.90 -> 60.77 | Improved |
| add backbone fine-tuning during incremental sessions (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | final session accuracy: 49.13 -> 50.14 | Improved |
| replace DeepInversion replay with entropy-regularized generator replay (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | average accuracy: 59.11 -> 60.77 | Improved |
| replace DeepInversion replay with entropy-regularized generator replay (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | final session accuracy: 48.48 -> 50.14 | Improved |
| add re-labeling and backbone fine-tuning to DeepInversion replay (ablation) | DeepInversion · CIFAR-100 FSCIL | average accuracy: 57.92 -> 59.11 | Improved |
| use replayed data (iCaRL*, NCM* re-implementations) instead of TOPIC without replay | NCM · CIFAR-100 FSCIL | average accuracy: 42.62 -> 58.60 | Improved |
| replay data quantity different from training data size (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | average accuracy | Hurt |
| use entropy regularization in generator on label distribution (ablation) | Entropy-regularized data-free replay · CIFAR-100 FSCIL | share of generated samples from few-shot classes | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- iCaRL* — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · better on 3 of 3 shared results
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
- Pre-trained Vision and Language Transformers Are Few-Shot Incremental Learners (CVPR 2024 · similarity 0.88)
- Autoencoder-Based Hybrid Replay for Class-Incremental Learning (ICML 2025 · similarity 0.88)
- Tensor Decomposition Based Memory-Efficient Incremental Learning (ICML 2025 · similarity 0.87)
- MIND: Multi-Task Incremental Network Distillation (AAAI 2024 · similarity 0.87)
- M2SD:Multiple Mixing Self-Distillation for Few-Shot Class-Incremental Learning (AAAI 2024 · similarity 0.86)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 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.
