Research map / Architectures and efficiency / RNNs and normalization
The Uncanny Similarity of Recurrence and Depth
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 distinct residual-block weights with recurrence and separate batch-normalization statistics per iteration | ResNet-18-style residual classifier · CIFAR-10 | accuracy: $94.69 \pm 0.03$ -> $93.99 \pm 0.10$ | Hurt |
| Replace distinct residual-block weights with recurrence and separate batch-normalization statistics per iteration | ResNet-18-style residual classifier · SVHN | accuracy: $95.48 \pm 0.11$ -> $96.04 \pm 0.08$ | Helped |
| Replace distinct residual-block weights with recurrence and separate batch-normalization statistics per iteration | ResNet-18-style residual classifier · EMNIST | accuracy: $87.71 \pm 0.09$ -> $87.42 \pm 0.19$ | Hurt |
| Increase effective depth by recurrent iterations instead of adding distinct layers | Image-classification networks · ImageNet, CIFAR-10, EMNIST, and SVHN | accuracy | No clear effect |
| Increase effective depth by recurrent iterations instead of adding distinct residual blocks | Residual maze solvers · Small, medium, and large mazes | accuracy | No clear effect |
| Add dilated convolutional filters (ablation) | Residual maze solvers · Small mazes | accuracy | Helped |
| Increase effective depth using recurrence or distinct layers | Residual maze solvers with dilated filters · Small mazes | accuracy: 75.07 -> 95.19 (recurrent); 75.59 -> 95.00 (feed-forward) | Helped |
Closest papers by meaning
- Task-Driven Convolutional Recurrent Models of the Visual System (NeurIPS 2018 · similarity 0.89)
- Learning Implicitly Recurrent CNNs Through Parameter Sharing (ICLR 2019 · similarity 0.88)
- Architectural Complexity Measures of Recurrent Neural Networks (NeurIPS 2016 · similarity 0.88)
- Hyperparameter Transfer Laws for Non-Recurrent Multi-Path Neural Networks (ICML 2026 · similarity 0.87)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · similarity 0.86)
- Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers (NeurIPS 2025 · 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.