Research map / Architectures and efficiency / Neural architecture search
Neural Architecture 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 |
|---|---|---|---|
| Add skip connections to the CIFAR-10 architecture (ablation) | Neural Architecture Search v1 · CIFAR-10 test | error rate (%): 7.97 -> 5.50 | Helped |
| Densely connect all layers with skip connections (ablation) | Neural Architecture Search v1 · CIFAR-10 test | error rate (%): 5.50 -> 5.56 | Hurt |
| Allow the controller to predict strides (ablation) | Neural Architecture Search · CIFAR-10 test | error rate (%): 5.50 -> 6.01 | Hurt |
| Add two pooling layers to the search space (ablation) | Neural Architecture Search · CIFAR-10 test | error rate (%): 6.01 -> 4.47 | Helped |
| Add 40 filters to each layer of the max-pooling architecture (ablation) | Neural Architecture Search v3 max pooling · CIFAR-10 test | error rate (%): 4.47 -> 3.65 | Helped |
| Use shared embeddings with the discovered recurrent cell (ablation) | Neural Architecture Search with base 8 · Penn Treebank test | perplexity: 67.9 -> 64.0 | Helped |
| Transfer the discovered cell to character language modeling (ablation) | Two layer LSTM · PTB character modeling test | bits per character: 1.243 -> 1.228 | Helped |
| Increase the transferred new-cell model from 6.57M to 16.28M parameters (ablation) | Two Layer with New Cell · PTB character modeling test | bits per character: 1.228 -> 1.214 | Helped |
| Add max as a combination function and sin as an activation (ablation) | Neural Architecture Search recurrent-cell search space · Penn Treebank | perplexity | No clear effect |
| Use policy gradient rather than random search (ablation) | Neural Architecture Search · Penn Treebank | perplexity | Helped |
Closest papers by meaning
- Neural Architecture Optimization (NeurIPS 2018 · similarity 0.88)
- Neural Optimizer Search with Reinforcement Learning (ICML 2017 · similarity 0.88)
- Designing Neural Network Architectures using Reinforcement Learning (ICLR 2017 · similarity 0.86)
- Real-Time Recurrent Learning using Trace Units in Reinforcement Learning (NeurIPS 2024 · similarity 0.85)
- Exploring the Promise and Limits of Real-Time Recurrent Learning (ICLR 2024 · similarity 0.85)
- Reinforcement Learning Finetunes Small Subnetworks in Large Language Models (NeurIPS 2025 · 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.