Research map / Architectures and efficiency / RNNs and normalization
Do Neural Operators Forget Geometry? The Forgetting Hypothesis in Deep Operator Learning
Yanming Xia, Angelica I Aviles-Rivero
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 geometry memory injection (Late) | FNO · LDC-NS | relative L2 error: 3.61e−1 -> 1.21e−1 | Improved |
| Add geometry memory injection (Early) | Transolver · AirfRANS | relative L2 error: 3.28e−2 -> 1.40e−2 | Improved |
| Add geometry memory injection (All layers) | Transolver · LDC-NSHT | relative L2 error: 2.73e−1 -> 2.30e−1 | Improved |
| Add geometry memory injection | LNO · LDC-NSHT | relative L2 error: 2.36e−1 -> 2.29e−1 | No clear effect |
| Add geometry memory injection (L2) | LNO · Darcy | relative L2 error: 1.52e−1 -> 1.12e−1 | Improved |
| Inject geometry only at final layer L3 (ablation) | Transolver · LDC-NS | relative L2 error: 1.72e−1 -> 5.48e−1 | Hurt |
| Inject geometry only at final layer L3 (ablation) | FNO · LDC-NS | relative L2 error: 3.61e−1 -> 1.22e−1 | Improved |
| Inject geometry at final layer L3 (ablation) | Transolver · Darcy | relative L2 error: 1.16e−1 -> 9.11e−2 | Improved |
| Use DeepONet encoder instead of U-Net for geometry memory (ablation) | FNO · LDC-NS | relative L2 error: 1.16e−1 -> 1.69e−1 (U-Net vs DeepONet) | Hurt |
| Use DeepONet geometry encoder instead of no memory (ablation) | Transolver · LDC-NS | relative L2 error: 1.72e−1 -> 1.46e−1 | Improved |
| Use additive injection instead of FiLM at Late layers (ablation) | Transolver · LDC-NS | relative L2 error: 1.03e−1 -> 2.83e−1 | Hurt |
| Use concatenation instead of FiLM at Late layers (ablation) | Transolver · LDC-NS | relative L2 error: 1.03e−1 -> 5.20e−1 | Hurt |
Closest papers by meaning
- The Impact of Geometric Complexity on Neural Collapse in Transfer Learning (NeurIPS 2024 · similarity 0.86)
- Adaptive Coordinate Transforms for Neural Operators (NeurIPS 2026 · similarity 0.85)
- Geometry Aware Operator Transformer as an efficient and accurate neural surrogate for PDEs on arbitrary domains (NeurIPS 2025 · similarity 0.84)
- The Geometric Wall: Manifold Structure Predicts Layerwise Sparse Autoencoder Scaling Laws (NeurIPS 2026 · similarity 0.84)
- Reference Neural Operators: Learning the Smooth Dependence of Solutions of PDEs on Geometric Deformations (ICML 2024 · similarity 0.84)
- Geometric Inductive Biases of Deep Networks: The Role of Data and Architecture (ICLR 2025 · 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.
