Research map / Continual learning / Online and replay-based
IIRC: Incremental Implicitly-Refined Classification
Mohamed Abdelsalam, Mojtaba Faramarzi, Shagun Sodhani, Sarath Chandar
Check 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 |
|---|---|---|---|
| use iCaRL-CNN/iCaRL-norm instead of other baselines | lifelong learners · IIRC-CIFAR/ImageNet | pw-JS | Improved |
| use cosine normalization (iCaRL-norm) instead of iCaRL-CNN | iCaRL · IIRC-ImageNet | pw-JS | Improved |
| learn the scalar eta instead of fixing it (ablation) | iCaRL-norm · IIRC | accuracy | Improved |
| use A-GEM instead of vanilla ER | ER · IIRC-CIFAR | pw-JS | Hurt |
| keep all old samples but with incomplete information (ER-infinite) instead of complete information (incremental joint) | replay · IIRC | pw-JS | Hurt |
| measure with pw-JS instead of JS | iCaRL family · IIRC-CIFAR | metric score | Hurt |
| train on more subsequent tasks | ER, LUCIR baselines · IIRC | per-task performance | Hurt |
| use LUCIR instead of iCaRL-norm for superclass association | class incremental · IIRC-CIFAR | superclass labelling | Mixed |
| use ER with buffer of 100 samples per class (ablation) | ER · IIRC-ImageNet | average performance | No clear effect |
| vary replay buffer size per class (ablation) | ER · IIRC-CIFAR/ImageNet | average performance | Mixed |
Closest papers by meaning
- Class-Incremental Learning With Strong Pre-Trained Models (CVPR 2022 · similarity 0.88)
- Bring Evanescent Representations to Life in Lifelong Class Incremental Learning (CVPR 2022 · similarity 0.88)
- Class-Incremental Learning: Survey and Performance Evaluation on Image Classification (TPAMI 2022 · similarity 0.88)
- Generative Multi-modal Models are Good Class Incremental Learners (CVPR 2024 · similarity 0.87)
- Specifying What You Know or Not for Multi-Label Class-Incremental Learning (AAAI 2025 · similarity 0.86)
- Navigating Semantic Drift in Task-Agnostic Class-Incremental Learning (ICML 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.
