Research map / Architectures and efficiency / RNNs and normalization
RNNs of RNNs: Recursive Construction of Stable Assemblies of Recurrent Neural Networks
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 |
|---|---|---|---|
| constrain inter-subnetwork connections to preserve contraction (ablation) | Sparse Combo Net · permuted seqMNIST, 24 × 32 network | test accuracy: 47.0% -> 96.7% | Helped |
| increase the number of modules from one to two (ablation) | Sparse Combo Net · permuted seqMNIST, 32 units per module | test accuracy: 37.1% -> 89.1% | Helped |
| increase the number of modules from one to two (ablation) | SVD Combo Net · permuted seqMNIST, 32 units per module | test accuracy: 61.8% -> 92.9% | Helped |
| reduce subnetwork sparsity from 26.5% to 10% (ablation) | Sparse Combo Net, 11 × 32 network · permuted seqMNIST | test accuracy: >5 percentage point | Helped |
| vary subnetwork sparsity; 3.3% sparsity performed best among tested 1%, 3.3%, 5%, and 10% levels (ablation) | Sparse Combo Net, 16 × 32 network · permuted seqMNIST and seqCIFAR10 | test accuracy | Mixed |
| decrease the magnitude of nonzero subnetwork weights while holding sparsity fixed (ablation) | Sparse Combo Net · permuted seqMNIST | test accuracy | Hurt |
| vary modularity while holding total size at 352 units (ablation) | Sparse Combo Net · permuted seqMNIST | test accuracy | Mixed |
| increase the number of neurons (ablation) | Sparse Combo Net and SVD Combo Net · permuted seqMNIST | test accuracy: saturating at ∼ 97% for Sparse Combo Net and ∼ 95% for SVD Combo Net | Helped |
Closest papers by meaning
- Learning Dynamics of RNNs in Closed-Loop Environments (NeurIPS 2025 · similarity 0.90)
- Unconditional stability of a recurrent neural circuit implementing divisive normalization (NeurIPS 2024 · similarity 0.90)
- Recurrent neural network dynamical systems for biological vision (NeurIPS 2024 · similarity 0.89)
- Measuring and Controlling Solution Degeneracy across Task-Trained Recurrent Neural Networks (NeurIPS 2025 · similarity 0.88)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.88)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.88)
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.