Research map / Architectures and efficiency / RNNs and normalization
Weight Agnostic Neural Networks
Adam Gaier, David Ha
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 |
|---|---|---|---|
| Search architectures with a single shared weight instead of using a fixed topology | CartPoleSwingUp · CartPoleSwingUp | cumulative reward with random shared weight: 7 ± 2 -> 515 ± 58 | Improved |
| Search architectures with a single shared weight instead of using a fixed topology | BipedalWalker-v2 · BipedalWalker-v2 | cumulative reward with random shared weight: -107 ± 12 -> 51 ± 108 | Improved |
| Search architectures with a single shared weight instead of using a fixed topology | CarRacing-v0 · CarRacing-v0 | cumulative reward with random shared weight: -85 ± 27 -> 375 ± 177 | Improved |
| Tune the shared weight after WANN architecture search | WANN · CartPoleSwingUp | cumulative reward: 515 ± 58 -> 723 ± 16 | Improved |
| Tune the shared weight after WANN architecture search | WANN · BipedalWalker-v2 | cumulative reward: 51 ± 108 -> 261 ± 58 | Improved |
| Tune the shared weight after WANN architecture search | WANN · CarRacing-v0 | cumulative reward: 375 ± 177 -> 608 ± 161 | Improved |
| Use a fixed single weight value rather than training across a range of values (ablation) | WANN search · continuous control tasks | robust task performance | Hurt |
| Use independently random weights for each connection rather than one shared weight (ablation) | WANN search · continuous control tasks | task performance | Hurt |
| Add Gaussian noise to the shared weight values (ablation) | WANN · CartPoleSwingUp | cumulative reward | No clear effect |
| Add very high levels of Gaussian noise to shared weight values (ablation) | WANN · CartPoleSwingUp | cumulative reward | Hurt |
| Fine-tune WANN weights with backpropagation rather than black-box optimization (ablation) | MNIST WANN · MNIST | accuracy | Hurt |
Closest papers by meaning
- Distribution Consistent Neural Architecture Search (CVPR 2022 · similarity 0.86)
- A Modern Self-Referential Weight Matrix That Learns to Modify Itself (ICML 2022 · similarity 0.85)
- Learning Useful Representations of Recurrent Neural Network Weight Matrices (ICML 2024 · similarity 0.85)
- Stable Gradients for Stable Learning at Scale in Deep Reinforcement Learning (NeurIPS 2025 · similarity 0.84)
- The Effect of Weight Precision on the Neuron Count in Deep ReLU Networks (ICML 2024 · similarity 0.84)
- Expressivity of Neural Networks with Random Weights and Learned Biases (ICLR 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.
