Research map / Continual learning / Class-incremental
Overcoming Catastrophic Forgetting With Unlabeled Data in the Wild
Kibok Lee, Kimin Lee, Jinwoo Shin, Honglak Lee
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 unlabeled external data | GD · CIFAR-100 (task size 10) | ACC: 65.0 -> 68.1 | Improved |
| add unlabeled external data | GD · CIFAR-100 (task size 10) | FGT: 12.1 -> 7.7 | Improved |
| add unlabeled external data | E2E · ImageNet 32x32 (task size 10) | ACC: 50.8 -> 52.8 | Improved |
| add current-task teacher C to previous model P (ablation) | GD · CIFAR-100 (task size 10) | ACC: 62.9 -> 67.0 | Improved |
| add ensemble reference Q to P and C (ablation) | GD · CIFAR-100 (task size 10) | ACC: 67.0 -> 68.1 | Improved |
| add ensemble reference Q to P and C (ablation) | GD · CIFAR-100 (task size 10) | FGT: 10.7 -> 7.7 | Improved |
| train C by distillation (dst) instead of direct hard-label loss (cls) (ablation) | GD · CIFAR-100 (task size 10) | ACC: 62.9 -> 66.2 | Improved |
| add confidence loss to teacher C (ablation) | GD · CIFAR-100 (task size 10) | ACC: 66.2 -> 67.0 | Improved |
| add fine-tuning with data weighting (FT-DW) (ablation) | GD · CIFAR-100 (task size 10) | FGT: 11.5 -> 7.7 | Improved |
| fine-tuning with data weighting instead of FT-DSet (ablation) | GD · CIFAR-100 (task size 10) | ACC: 67.2 -> 68.1 | Improved |
| sample random OOD data in addition to previous-task-like data (ablation) | GD · CIFAR-100 (task size 10) | ACC: 66.0 -> 68.1 | Improved |
| sample OOD data by previous-model prediction instead of randomly (ablation) | GD · CIFAR-100 (task size 10) | ACC: 68.1 -> 65.7 | Hurt |
Closest papers by meaning
- A comprehensive, application-oriented study of catastrophic forgetting in DNNs (ICLR 2019 · similarity 0.89)
- Learn to Grow: A Continual Structure Learning Framework for Overcoming Catastrophic Forgetting (ICML 2019 · similarity 0.89)
- Maintaining Discrimination and Fairness in Class Incremental Learning (CVPR 2020 · similarity 0.88)
- Measuring Catastrophic Forgetting in Neural Networks (AAAI 2018 · similarity 0.88)
- Decoupled Distillation to Erase: A General Unlearning Method for Any Class-centric Tasks (CVPR 2025 · similarity 0.87)
- ZeroFlow: Overcoming Catastrophic Forgetting is Easier than You Think (ICML 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.
