Research map / Architectures and efficiency / CNN design
Convolutional Neural Networks with Gated Recurrent Connections
Jianfeng Wang, Xiaolin Hu
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 |
|---|---|---|---|
| add gates on recurrent connections (RCNN -> original GRCNN), T=5, tied weights (ablation) | RCNN · CIFAR-10 | test error (%): 5.88 -> 5.57 | Improved |
| add gates, T=2, tied weights (ablation) | RCNN · CIFAR-10 | test error (%): 5.81 -> 5.92 | Hurt |
| add accumulated contributions over iterations (original -> improved GRCNN), T=5 untied (ablation) | GRCNN · CIFAR-10 | test error (%): 5.23 -> 5.16 | Improved |
| increase iterations T from 2 to 5 in each GRCL, tied weights (ablation) | original GRCNN · CIFAR-10 | test error (%): 5.92 -> 5.57 | Improved |
| increase iterations T from 2 to 5, tied weights (ablation) | improved GRCNN · CIFAR-10 | test error (%): 5.83 -> 5.49 | Improved |
| increase iterations T from 2 to 5, tied weights (ablation) | RCNN · CIFAR-10 | test error (%): 5.65 -> 5.88 | Hurt |
| untie weights across iterations, T=5 (ablation) | original GRCNN · CIFAR-10 | test error (%): 5.57 -> 5.23 | Improved |
| add selective kernel convolution to GRCNN-110 | GRCNN-110 · CIFAR-10 | test error (%): 3.82 -> 3.42 | Improved |
| replace SK block in SKNet-110 with gated recurrent SK-GRCNN-110 | SKNet-110 · CIFAR-100 | test error (%): 18.34 -> 18.18 | Improved |
| replace ResNet-50 with GRCNN-55 | ResNet-50 · ImageNet val | top-1 error (%): 24.70 -> 23.12 | Improved |
| add deformable v2 convolution to GRCNN-55 | GRCNN-55 · ImageNet val | top-1 error (%): 23.12 -> 21.63 | Improved |
| add recurrent iterations T=1 to T=3 with gates (ablation) | GRCNN · IIIT5K | accuracy (%): 78.63 -> 81.01 | Improved |
Closest papers by meaning
- Gated Recurrent Convolution Neural Network for OCR (NeurIPS 2017 · similarity 0.91)
- Task-Driven Convolutional Recurrent Models of the Visual System (NeurIPS 2018 · similarity 0.88)
- Learning long-range spatial dependencies with horizontal gated recurrent units (NeurIPS 2018 · similarity 0.88)
- Recurrence along Depth: Deep Convolutional Neural Networks with Recurrent Layer Aggregation (NeurIPS 2021 · similarity 0.87)
- Learning Implicitly Recurrent CNNs Through Parameter Sharing (ICLR 2019 · similarity 0.87)
- Recurrent neural network dynamical systems for biological vision (NeurIPS 2024 · similarity 0.85)
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.
