Research map / Continual learning / Class-incremental
Learnable Distribution Calibration for Few-Shot Class-Incremental Learning
Binghao Liu, Boyu Yang, Lingxi Xie, Ren Wang, Qi Chuan Tian, Qixiang Ye
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 Gaussian sampler to baseline (ablation) | LDC · CUB200 | accuracy (%): 51.86 -> 54.58 | Improved |
| add recurrent calibration module to Gaussian sampler (ablation) | LDC · CUB200 | accuracy (%): 54.58 -> 61.58 | Improved |
| use KL divergence instead of EMD matching loss (ablation) | LDC · CUB200 | accuracy (%): 59.21 -> 61.58 | Improved |
| use KL divergence instead of JS divergence (ablation) | LDC · CUB200 | accuracy (%): 58.68 -> 61.58 | Improved |
| use KL divergence instead of Hellinger distance (ablation) | LDC · CUB200 | accuracy (%): 58.32 -> 61.58 | Improved |
| use learned covariance mapping instead of empirical strategy (ablation) | LDC · mini-ImageNet | convergence epoch: 312 -> 206 | Improved |
| use learned covariance mapping instead of empirical strategy (ablation) | LDC · CIFAR100 | generalization accuracy: 27.55% -> 31.36% | Improved |
| replace Free-Lunch with LDC on regular samples | Free-Lunch · mini-ImageNet FSCIL normal cases | PD: 25.62 -> 22.51 | Improved |
| replace Free-Lunch with LDC on outlier few-shot samples | Free-Lunch · mini-ImageNet FSCIL outlier cases | PD: 30.07 -> 25.32 | Improved |
| replace Free-Lunch with LDC | Free-Lunch · CUB200 FSCIL | accuracy last session (%): 55.68 -> 61.58 | Improved |
| replace Free-Lunch with LDC in regular few-shot | Free-Lunch · CUB 5-way 1-shot | accuracy (%): 79.56 -> 81.51 | Improved |
| replace CEC with LDC | CEC · mini-ImageNet FSCIL | accuracy last session (%): 47.63% -> 54.46% | 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.
- Meta-FSCIL — MetaFSCIL: A Meta-Learning Approach for Few-Shot Class Incremental Learning (CVPR 2022) · better on 1 of 1 shared result
- iCaRL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · 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
- Few-Shot Class-Incremental Learning via Training-Free Prototype Calibration (NeurIPS 2023 · similarity 0.91)
- Generalized and Incremental Few-Shot Learning by Explicit Learning and Calibration Without Forgetting (ICCV 2021 · similarity 0.91)
- Rethinking Few-shot Class-incremental Learning: Learning from Yourself (ECCV 2024 · similarity 0.90)
- Bridging the Gap Between Few-Shot and Many-Shot Learning via Distribution Calibration (TPAMI 2021 · similarity 0.90)
- Enhancing Few-Shot Class-Incremental Learning via Training-Free Bi-Level Modality Calibration (CVPR 2025 · similarity 0.88)
- Adaptive Decision Boundary for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.87)
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.
