Research map / Architectures and efficiency / CNN design
Learning Rate Curriculum
Florinel-Alin Croitoru, Nicolae-Cătălin Ristea, Radu Tudor Ionescu, Nicu Sebe
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 LeRaC layer-wise learning rate curriculum | ResNet-18 · ImageNet (from scratch) | accuracy: 68.44 -> 71.96 | Improved |
| add LeRaC | ResNet-18 · Food-101 (from scratch) | accuracy: 68.31 -> 69.57 | Improved |
| add LeRaC | CvT-13 (from scratch) · Food-101 | accuracy: 39.17 -> 41.42 | Improved |
| add LeRaC | CvT-13 (pre-trained) · CIFAR-100 | accuracy: 77.80 -> 78.93 | Improved |
| add LeRaC | BERT · BoolQ | accuracy: 74.12 -> 75.55 | Improved |
| add LeRaC | YOLOv5 · PASCAL VOC 2007+2012 | mAP: 0.832 -> 0.846 | Improved |
| add Curriculum by Smoothing (CBS) | CvT-13 (pre-trained) · CIFAR-10 | accuracy: 93.56 -> 85.85 | Hurt |
| add CBS | Wide-ResNet-50 · Tiny ImageNet (from scratch) | accuracy: 55.97 -> 48.30 | Hurt |
| add CBS | ResNet-18 · ImageNet (from scratch) | accuracy: 68.44 -> 71.02 | Improved |
| use exponential instead of linear learning rate update in LeRaC (ablation) | CvT-13 (from scratch) · CIFAR-10 | accuracy: 72.49 -> 72.90 | Improved |
| use exponential instead of logarithmic learning rate update in LeRaC (ablation) | CvT-13 (from scratch) · CIFAR-10 | accuracy: 72.14 -> 72.90 | Improved |
| use curriculum (lower rates for deeper layers) instead of anti-LeRaC (ablation) | ResNet-18 · CIFAR-100 | accuracy: 71.24 -> 72.72 | Improved |
Closest papers by meaning
- On The Power of Curriculum Learning in Training Deep Networks (ICML 2019 · similarity 0.88)
- A Survey on Curriculum Learning (TPAMI 2021 · similarity 0.87)
- Curriculum Learning by Transfer Learning: Theory and Experiments with Deep Networks (ICML 2018 · similarity 0.86)
- EfficientTrain: Exploring Generalized Curriculum Learning for Training Visual Backbones (ICCV 2023 · similarity 0.86)
- CurBench: Curriculum Learning Benchmark (ICML 2024 · similarity 0.84)
- Dual-Level Curriculum Meta-Learning for Noisy Few-Shot Learning Tasks (AAAI 2024 · similarity 0.83)
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.
