Research map / Architectures and efficiency / RNNs and normalization
On The Power of Curriculum Learning in Training Deep Networks
Read the paperCheck 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 |
|---|---|---|---|
| Order examples from easy to hard using transfer scores and fixed exponential pacing | Moderate CNN · CIFAR-100 small mammals, CIFAR-10, CIFAR-100, and ImageNet cats subset | test accuracy | Helped |
| Order examples from easy to hard using transfer scores and varied exponential pacing | VGG-based network · CIFAR-10 and CIFAR-100 | test accuracy | Helped |
| Use self-taught target-hypothesis scores for bootstrapping (ablation) | Moderate CNN · CIFAR-100 small mammals | test accuracy | Helped |
| Replace exponential pacing with single-step pacing (ablation) | Transfer-scored curriculum · CIFAR-100 small mammals | test accuracy | Helped |
| Vary the step lengths in exponential pacing (ablation) | Transfer-scored curriculum · CIFAR-100 small mammals | test accuracy | Helped |
| Score examples by current-hypothesis loss rather than target-hypothesis loss (ablation) | Curriculum training · CIFAR-100 small mammals | test accuracy | Hurt |
| Present harder examples before easier examples (ablation) | Moderate CNN · CIFAR-100 small mammals | test accuracy | Mixed |
| Randomize example difficulty scores (ablation) | Moderate CNN · CIFAR-100 small mammals | test accuracy | No clear effect |
| Repeat bootstrapping to generate successive scoring functions (ablation) | Curriculum by bootstrapping · CIFAR-100 small mammals | test accuracy | Hurt |
| Use cyclic rather than exponential learning-rate scheduling (ablation) | Curriculum and vanilla training · CIFAR-10 and CIFAR-100 | test accuracy | Helped |
| Use transfer scoring on a harder classification task | Curriculum learning · CIFAR-10 versus CIFAR-100 | test accuracy | Mixed |
Closest papers by meaning
- Curriculum Learning by Transfer Learning: Theory and Experiments with Deep Networks (ICML 2018 · similarity 0.93)
- Curriculum By Smoothing (NeurIPS 2020 · similarity 0.89)
- Curriculum Learning by Dynamic Instance Hardness (NeurIPS 2020 · similarity 0.88)
- MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels (ICML 2018 · similarity 0.88)
- When Does Curriculum Learning Help? A Theoretical Perspective (NeurIPS 2025 · similarity 0.88)
- CurBench: Curriculum Learning Benchmark (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.