Research map / Continual learning / Class-incremental
iCaRL: Incremental Classifier and Representation 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 |
|---|---|---|---|
| add nearest-mean-of-exemplars classification instead of using network outputs (ablation) | iCaRL · iCIFAR-100, 2 classes per batch | average incremental accuracy: 36.6 -> 57.0 | Helped |
| add nearest-mean-of-exemplars classification instead of using network outputs (ablation) | iCaRL · iCIFAR-100, 10 classes per batch | average incremental accuracy: 59.3 -> 64.1 | Helped |
| add distillation loss during training (ablation) | iCaRL with exemplar classification · iCIFAR-100, 2 classes per batch | average incremental accuracy: 57.6 -> 57.0 | Hurt |
| add distillation loss during training (ablation) | iCaRL with exemplar classification · iCIFAR-100, 50 classes per batch | average incremental accuracy: 65.3 -> 68.6 | Helped |
| add exemplar images during representation learning (ablation) | LwF.MC · iCIFAR-100, 10 classes per batch | average incremental accuracy: 44.4 -> 58.1 | Helped |
| replace mean-of-exemplars classification with nearest-class-mean classification (ablation) | iCaRL · iCIFAR-100, 10 classes per batch | average incremental accuracy: 64.1 -> 64.5 | No clear effect |
Closest papers by meaning
- Class-Incremental Learning With Strong Pre-Trained Models (CVPR 2022 · similarity 0.90)
- Bring Evanescent Representations to Life in Lifelong Class Incremental Learning (CVPR 2022 · similarity 0.89)
- Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning (CVPR 2022 · similarity 0.88)
- A Model or 603 Exemplars: Towards Memory-Efficient Class-Incremental Learning (ICLR 2023 · similarity 0.88)
- eTag: Class-Incremental Learning via Embedding Distillation and Task-Oriented Generation (AAAI 2024 · similarity 0.87)
- Harnessing Neural Unit Dynamics for Effective and Scalable Class-Incremental Learning (ICML 2024 · similarity 0.85)
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.