Research map / Architectures and efficiency / RNNs and normalization
Integral Autoencoder Network for Discretization-Invariant Learning
Yong Zheng Ong, Zuowei Shen, Haizhao Yang
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 |
|---|---|---|---|
| use IAE-Net instead of dgFNO+ | Darcy problem in notched triangular domain · Darcy problem, triangular domain with notch (s=50) | L2 relative error (%): 7.821% -> 0.8094% | Improved |
| use IAE-Net instead of POD-DeepONet | Darcy problem in notched triangular domain · Darcy problem, triangular domain with notch (s=50) | L2 relative error (%): 1.00% -> 0.8094% | Improved |
| test IAE-Net at higher resolutions than trained | IAE-Net · Darcy problem, triangular domain with notch | L2 relative error (%): 0.8094% -> 0.8317% | No clear effect |
| add randomized data augmentation across resolutions to IAE-Net (ablation) | IAE-Net · Burgers equation benchmark (nu=1e-1) | L2 relative error across resolutions | Improved |
| use dense skip connections instead of ResNet skip connections in IAE-Net (ablation) | IAE-Net · Burgers equation benchmark | L2 relative error: almost 2 times better than the ResNet counterpart | Improved |
| use ResNet skip connections instead of dense connections in IAE-Net (ablation) | IAE-Net · Darcy benchmark | L2 relative error | Hurt |
| use multiple IAE blocks (4) instead of one block with deeper networks (ablation) | IAE-Net · Burgers equation benchmark (s=1024) | L2 relative error: reducing error by 2 times | Improved |
| increase number of IAE blocks to L=5 (ablation) | IAE-Net · Burgers equation benchmark (s=1024) | L2 relative error | Mixed |
| use IAE-Net instead of FNO, FT and GT | Burgers equation (nu=1e-1) · Burgers equation | L2 relative error: almost 3 times better | Improved |
| use IAE-Net instead of FNO, FT and GT | scattering problem · Scattering forward and inverse | L2 relative error: more than 10 times higher error for baselines | Improved |
| use IAE-Net instead of FNO, FT and GT | CT image denoising · CT denoising (30 view, 5% noise) | L2 relative error: about 3 times lower error | Improved |
| train on resolutions sequentially instead of jointly with summed gradients (ablation) | IAE-Net · Scattering, mixed resolutions | test error | Hurt |
Closest papers by meaning
- Integral Neural Networks (CVPR 2023 · similarity 0.84)
- Deep Neural Network Approximation of Invariant Functions through Dynamical Systems (JMLR 2024 · similarity 0.82)
- MIRE: Matched Implicit Neural Representations (CVPR 2025 · similarity 0.82)
- Training Integrable Parameterizations of Deep Neural Networks in the Infinite-Width Limit (JMLR 2024 · similarity 0.82)
- On Random Deep Weight-Tied Autoencoders: Exact Asymptotic Analysis, Phase Transitions, and Implications to Training (ICLR 2019 · similarity 0.82)
- Axial Neural Networks for Dimension-Free Foundation Models (NeurIPS 2025 · similarity 0.81)
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.
