Research map / Optimization / Optimizers
Neural Optimizer Search with Reinforcement Learning
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 |
|---|---|---|---|
| Replace Adam with PowerSign | GNMT · WMT 2014 English-German | Test BLEU: 24.5 -> 25.0 | Helped |
| Replace RMSProp with PowerSign-cd | Mobile-sized NASNet-A · ImageNet | Top-1 accuracy: 73.5 -> 73.9 | Helped |
| Replace RMSProp with AddSign-ld | Mobile-sized NASNet-A · ImageNet | Top-1 accuracy: 73.5 -> 73.8 | Helped |
| Use PowerSign instead of Adam | 10M single-layer LSTM · Penn Treebank | Test perplexity: 63.3 -> 64.8 | Hurt |
| Use AddSign instead of Adam | 10M single-layer LSTM · Penn Treebank | Test perplexity: 63.3 -> 65.2 | Hurt |
| Add a linear internal decay to PowerSign (ablation) | PowerSign on Wide ResNet · CIFAR-10 (no learning rate decay) | Best Test accuracy: 93.0 -> 93.6 | Helped |
| Add cyclical internal decay to PowerSign (ablation) | PowerSign on Wide ResNet · CIFAR-10 (no learning rate decay) | Best Test accuracy: 93.0 -> 93.7 | Helped |
| Add restart internal decay with n=20 to PowerSign (ablation) | PowerSign on Wide ResNet · CIFAR-10 (no learning rate decay) | Best Test accuracy: 93.0 -> 94.4 | Helped |
| Add restart internal decay with n=20 to AddSign (ablation) | AddSign on Wide ResNet · CIFAR-10 (no learning rate decay) | Best Test accuracy: 93.0 -> 94.4 | Helped |
| Use noisy linear cosine instead of cosine learning-rate decay (ablation) | AddSign-ld on Wide ResNet · CIFAR-10 | Best Test accuracy: 95.4 -> 95.5 | No clear effect |
| Replace Adam with PowerSign | GNMT · WMT 2014 English-German | Train perplexity: 1.49 -> 1.39 | Helped |
Closest papers by meaning
- Towards Learned Optimization Free Lunch (ICLR 2026 · similarity 0.88)
- Learning to Learn without Gradient Descent by Gradient Descent (ICML 2017 · similarity 0.88)
- Neural Architecture Search with Reinforcement Learning (ICLR 2017 · similarity 0.88)
- Reinforced Genetic Algorithm Learning for Optimizing Computation Graphs (ICLR 2020 · similarity 0.88)
- Investigation into Training Dynamics of Learned Optimizers (Student Abstract) (AAAI 2024 · similarity 0.87)
- DeepSearch: Overcome the Bottleneck of Reinforcement Learning with Verifiable Rewards via Monte Carlo Tree Search (ICLR 2026 · 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.