Research map / Graph learning / GNN architectures
Learning Connectivity of Neural Networks from a Topological Perspective
Kun Yuan, Quanquan Li, Jing Shao, Junjie Yan
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 learnable edge weights alpha to the complete graph (ablation) | TopoNet complete · ImageNet val | Top-1 accuracy: 77.24 -> 78.22 | Improved |
| add uniform sparsity constraint on alpha (ablation) | TopoNet complete + alpha · ImageNet val | Top-1 accuracy: 78.22 -> 78.46 | Improved |
| add adaptive (in-degree log-scaled) sparsity constraint on alpha (ablation) | TopoNet complete + alpha · ImageNet val | Top-1 accuracy: 78.22 -> 78.60 | Improved |
| replace uniform with adaptive sparsity (ablation) | TopoNet complete + alpha · ImageNet val | Top-1 accuracy: 78.46 -> 78.60 | Improved |
| learn connectivity on ResNet-110 (complete graph, interval 1) | ResNet-110 · CIFAR-100 val | Top-1 accuracy: 76.31 -> 78.54 | Improved |
| learn connectivity on ResNet-20 (smaller search space) | ResNet-20 · CIFAR-100 val | Top-1 accuracy: 69.01 -> 69.91 | Improved |
| learn connectivity on MobileNetV2-1.0 baseline depth | MobileNetV2-1.0 · ImageNet val | Top-1 accuracy: 72.60 -> 72.86 | Improved |
| learn connectivity on 6x deeper MobileNetV2 (larger search space) | MobileNetV2-1.0-6N · ImageNet val | Top-1 accuracy: 77.61 -> 78.36 | Improved |
| replace residual l=2 topology with learned complete topology with adaptive sparsity | TopoNet · COCO val2017 | AP: 40.74 -> 41.69 | Improved |
| replace ResNet-50 backbone with learned TopoNet backbone | FPN · COCO val2017 | AP: 36.42 -> 41.69 | Improved |
| reduce residual interval from 4 to 2 (denser connections) (ablation) | TopoNet residual · ImageNet val | Top-1 accuracy: 77.72 -> 78.26 | Improved |
| make the complete graph random with p=0.4 instead of fully connected (ablation) | TopoNet · ImageNet val | Top-1 accuracy: 77.24 -> 77.90 | Improved |
Closest papers by meaning
- Copresheaf Topological Neural Networks: A Generalized Deep Learning Framework (NeurIPS 2025 · similarity 0.87)
- A Topological Perspective on Demystifying GNN-Based Link Prediction Performance (ICLR 2024 · similarity 0.87)
- Input Space Mode Connectivity in Deep Neural Networks (ICLR 2025 · similarity 0.87)
- Topological Blindspots: Understanding and Extending Topological Deep Learning Through the Lens of Expressivity (ICLR 2025 · similarity 0.87)
- Topological Generalization Bounds for Discrete-Time Stochastic Optimization Algorithms (NeurIPS 2024 · similarity 0.86)
- Weisfeiler and Lehman Go Paths: Learning Topological Features via Path Complexes (AAAI 2024 · 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.
