Research map / Architectures and efficiency / RNNs and normalization
Scalable Learning to Optimize: A Learned Optimizer Can Train Big Models
Xuxi Chen, Tianlong Chen, Yu Cheng, Weizhu Chen, Ahmed Awadallah, Zhangyang Wang
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 tiny-subspace reparameterization to Momentum | Momentum, ResNet-18 · CIFAR-10 | convergence steps: 4316.8 -> 216.8 | Improved |
| add learned optimizer SL2O on top of subspace training | Momentum† ResNet-18 · CIFAR-10 | convergence steps: 216.8 -> 63.6 | Improved |
| add learned optimizer SL2O on top of subspace training | Momentum† ResNet-18 · CIFAR-100 | convergence steps: 283.2 -> 61.6 | Improved |
| add learned optimizer SL2O on top of subspace training | Momentum† VGG-16 · CIFAR-100 | convergence steps: 2104.8 -> 168.2 | Improved |
| add SL2O instead of Momentum | Momentum ResNet-18 · CIFAR-10 | test accuracy (%): 93.90 -> 93.61 | No clear effect |
| add SL2O instead of Momentum† | Momentum† VGG-16 · CIFAR-100 | test accuracy (%): 70.98 -> 71.02 | No clear effect |
| add subspace training to SGD | SGD VGG-16 · CIFAR-10 | test accuracy (%): 92.81 -> 89.03 | Hurt |
| add subspace training to SGD | SGD ResNet-20 · CIFAR-100 | test accuracy (%): 63.18 -> 66.64 | Improved |
| add SL2O instead of Momentum† | Momentum† ResNet-18 · ImageNet (40 epochs) | test accuracy (%): 68.83 -> 68.78 | No clear effect |
| meta-train SL2O on ResNet-18 instead of on ResNet-20 (ablation) | SL2O ResNet-20 · CIFAR-10 | test accuracy (%): 91.00 -> 90.96 | No clear effect |
| switch to Momentum after first 5k steps of SL2O | SL2O ViT-B/16 · CIFAR-10 | test accuracy: +0.12% | Improved |
| add SL2O instead of Momentum† 0.04 | Momentum† ViT-B/16 · CIFAR-10 | convergence steps: 602 -> 484 | Improved |
Closest papers by meaning
- Training Stronger Baselines for Learning to Optimize (NeurIPS 2020 · similarity 0.90)
- M-L2O: Towards Generalizable Learning-to-Optimize by Test-Time Fast Self-Adaptation (ICLR 2023 · similarity 0.90)
- Symbolic Learning to Optimize: Towards Interpretability and Scalability (ICLR 2022 · similarity 0.90)
- Learning Provably Improves the Convergence of Gradient Descent (NeurIPS 2025 · similarity 0.88)
- On the Parameterization of Second-Order Optimization Effective towards the Infinite Width (ICLR 2024 · similarity 0.86)
- Towards Robust Learning to Optimize with Theoretical Guarantees (CVPR 2024 · 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.
