Research map / Optimization / Implicit bias and generalization
Learning Discretized Neural Networks under Ricci Flow
Jun Chen, Hanwen Chen, Mengmeng Wang, Guang Dai, Ivor W. Tsang, Yong Liu
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 |
|---|---|---|---|
| replace Dorefa gradient with RF-DNN Ricci-flow gradient (ResNet20) (ablation) | binarized ResNet training with straight-through estimators · CIFAR-10 (1-bit weights, from scratch) | test accuracy (%): 88.28 -> 89.83 | Improved |
| replace MultiFCG gradient with RF-DNN (ResNet32) (ablation) | binarized ResNet training with straight-through estimators · CIFAR-10 (1-bit weights, from scratch) | test accuracy (%): 89.63 -> 90.75 | Improved |
| replace Dorefa gradient with RF-DNN (ResNet44) (ablation) | binarized ResNet training with straight-through estimators · CIFAR-10 (1-bit weights, from scratch) | test accuracy (%): 90.71 -> 91.63 | Improved |
| replace FCGrad gradient with RF-DNN (ResNet56) (ablation) | binarized ResNet training with straight-through estimators · CIFAR-100 (1-bit weights, from scratch) | test accuracy (%): 66.56 -> 68.56 | Improved |
| replace FCGrad gradient with RF-DNN (ResNet110) (ablation) | binarized ResNet training with straight-through estimators · CIFAR-100 (1-bit weights, from scratch) | test accuracy (%): 68.74 -> 69.20 | Improved |
| replace Dorefa gradient with RF-DNN (ResNet18) (ablation) | binarized ResNet training with straight-through estimators · ImageNet (1-bit weights, from scratch) | top-1 accuracy (%): 58.34 -> 60.83 | Improved |
| replace FCGrad gradient with RF-DNN (ResNet18) (ablation) | binarized ResNet training with straight-through estimators · ImageNet (1-bit weights, from scratch) | top-5 accuracy (%): 82.67 -> 83.54 | Improved |
| replace MultiFCG gradient with RF-DNN (ResNet44) (ablation) | binarized ResNet training with straight-through estimators · CIFAR-10 (1-bit weights, from scratch) | test accuracy (%): 90.54 -> 91.63 | Improved |
Closest papers by meaning
- Graph Neural Ricci Flow: Evolving Feature from a Curvature Perspective (ICLR 2025 · similarity 0.87)
- Neural Feature Geometry Evolves as Discrete Ricci Flow (ICML 2026 · similarity 0.86)
- A Riemannian Framework for Learning Reduced-order Lagrangian Dynamics (ICLR 2025 · similarity 0.84)
- Implicit regularization of deep residual networks towards neural ODEs (ICLR 2024 · similarity 0.84)
- RMLR: Extending Multinomial Logistic Regression into General Geometries (NeurIPS 2024 · similarity 0.84)
- Learning Theory of Transformers: Local-to-Global Approximation via Softmax Partition of Unity (NeurIPS 2026 · 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.
