Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Learning Across Scales---Multiscale Methods for Convolution 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 |
|---|---|---|---|
| Prolongate convolution kernels and classifier weights with multigrid operators instead of using the original kernels (ablation) | CNN trained at coarse resolution · MNIST validation, fine resolution | accuracy: $61.02$\% -> $91.02$\% | Helped |
| Restrict the convolution kernel when transferring the fine-trained network to coarse resolution (ablation) | CNN trained at fine resolution · MNIST validation, coarse resolution | accuracy: $84.09$\% -> $94.92$\% | Helped |
| Train on coarse images, then prolongate parameters and continue training at fine resolution | ResNet-34 · ImageNet-10 | training and validation error | Helped |
| Initialize deeper networks by prolongating parameters from shallower networks | MNIST CNNs, depths 2 to 64 layers · MNIST | validation accuracy | Helped |
| Initialize deeper networks by prolongating parameters from shallower networks instead of random initialization | MNIST CNNs, depths 2 to 64 layers · MNIST | iterations to reach comparable accuracy | Helped |
| Train first at 112 × 112, then prolongate weights and train at 224 × 224 | ResNet-34 · ImageNet-10 | training time | Helped |
Closest papers by meaning
- ScaleNet - Improve CNNs through Recursively Rescaling Objects (AAAI 2020 · similarity 0.87)
- Multigrid Neural Architectures (CVPR 2017 · similarity 0.87)
- The Multiscale Single-Index Model: A Toy Model for Hierarchical Feature Learning (NeurIPS 2026 · similarity 0.87)
- Multiscale Low-Frequency Memory Network for Improved Feature Extraction in Convolutional Neural Networks (AAAI 2024 · similarity 0.87)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · similarity 0.86)
- Depthwise Hyperparameter Transfer in Residual Networks: Dynamics and Scaling Limit (ICLR 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.