Research map / Continual learning / Class-incremental
S3C: Self-Supervised Stochastic Classifiers for Few-Shot Class-Incremental Learning
Jayateja Kalla, Soma Biswas
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 self-supervision (rotation) to linear classifier (ablation) | linear classifier · CIFAR100 | harmonic mean after final task (%): 26.76 -> 41.80 | Improved |
| add self-supervision (rotation) to linear classifier (ablation) | linear classifier · CIFAR100 | top1 accuracy after final task (%): 48.98 -> 53.55 | Improved |
| add self-supervision (rotation) to linear classifier (ablation) | linear classifier · CIFAR100 | base task accuracy (%): 74.70 -> 76.14 | Improved |
| replace linear classifier with stochastic classifier (ablation) | S3C with self-supervision · CIFAR100 | harmonic mean after final task (%): 41.80 -> 45.22 | Improved |
| replace linear classifier with stochastic classifier (ablation) | S3C with self-supervision · CIFAR100 | top1 accuracy after final task (%): 53.55 -> 53.96 | No clear effect |
| replace linear classifier with stochastic classifier (ablation) | S3C with self-supervision · CIFAR100 | base task accuracy (%): 76.14 -> 78.03 | Improved |
| S3C instead of CEC | CEC · CIFAR100 | top1 accuracy (%): 49.14 -> 53.96 | Improved |
| S3C instead of CEC | CEC · miniImageNet | top1 accuracy (%): 47.63 -> 52.14 | Improved |
| S3C instead of CEC | CEC · CUB200 | accuracy at session 10 (%): 52.28 -> 58.95 | Improved |
| S3C instead of CEC | CEC · CUB200 | performance dropping rate (%): 23.57 -> 20.83 | Improved |
| S3C instead of CEC | CEC · miniImageNet | performance dropping rate: 0.35% | Hurt |
| use only 40 base classes (FSCIL-lb), S3C instead of CEC | CEC · CIFAR100 (40 base classes) | top1 accuracy: 5.29% relative top1, 18.83% HM | 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 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 Incremental Learning With Continually Evolved Classifiers (CVPR 2021 · similarity 0.93)
- Few-Shot Class-Incremental Learning for Classification and Object Detection: A Survey (TPAMI 2025 · similarity 0.91)
- Tripartite Weight-Space Ensemble for Few-Shot Class-Incremental Learning (CVPR 2025 · similarity 0.90)
- Adaptive Decision Boundary for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.89)
- Pseudo Informative Episode Construction for Few-Shot Class-Incremental Learning (AAAI 2025 · similarity 0.88)
- M2SD:Multiple Mixing Self-Distillation for Few-Shot Class-Incremental Learning (AAAI 2024 · 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.
