Research map / Label-efficient and robust learning / Knowledge distillation
Adaptive Curriculum Learning
Yajing Kong, Liu Liu, Jun Wang, Dacheng Tao
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 difficulty-score adaptation (with KL loss) (ablation) | Adaptive CL · CIFAR-100 | validation accuracy: 45.49 -> 46.08 | Improved |
| add KL divergence loss (with adaptation) (ablation) | Adaptive CL · CIFAR-100 | validation accuracy: 45.88 -> 46.08 | Improved |
| add difficulty-score adaptation (with KL loss) (ablation) | Adaptive CL · HE devices | validation accuracy: 72.25 -> 74.06 | Improved |
| add KL divergence loss (with adaptation) (ablation) | Adaptive CL · HE devices | validation accuracy: 73.64 -> 74.06 | Improved |
| replace fixed curriculum with Adaptive CL | FCL · CIFAR-100 | validation accuracy: 45.25 -> 46.08 | Improved |
| replace random sampling with Adaptive CL | Vanilla · CIFAR-100 | validation accuracy: 42.16 -> 46.08 | Improved |
| use self-paced learning instead of vanilla sampling | Vanilla · CIFAR-100 | validation accuracy: 42.16 -> 39.70 | Hurt |
| use focal loss instead of vanilla sampling | Vanilla · CIFAR-100 | validation accuracy: 42.16 -> 42.34 | No clear effect |
| use VGG-16-based initial difficulty score: Adaptive CL instead of FCL | FCL · HE devices | validation accuracy: 73.32 -> 74.17 | Improved |
| use VGG-16-based initial difficulty score: Adaptive CL instead of FCL | FCL · MS mammals | validation accuracy: 75.34 -> 75.84 | Improved |
| use VGG-16-based initial difficulty score: Adaptive CL instead of FCL | FCL · Flowers | validation accuracy: 68.90 -> 69.45 | Improved |
| use negative alpha (adaptation) instead of alpha=0 (ablation) | Adaptive CL w/o KL · HE devices | validation accuracy: about 2% | Improved |
Closest papers by meaning
- A Survey on Curriculum Learning (TPAMI 2021 · similarity 0.91)
- Dual-Level Curriculum Meta-Learning for Noisy Few-Shot Learning Tasks (AAAI 2024 · similarity 0.87)
- Learning to Solve Complex Problems via Dataset Decomposition (NeurIPS 2025 · similarity 0.86)
- Bandit Guided Submodular Curriculum for Adaptive Subset Selection (NeurIPS 2025 · similarity 0.86)
- AdaCuRL: Adaptive Curriculum Reinforcement Learning with Invalid Sample Mitigation and Historical Revisiting (AAAI 2026 · similarity 0.86)
- CurBench: Curriculum Learning Benchmark (ICML 2024 · similarity 0.84)
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.
