Research map / Architectures and efficiency / Attention and transformers
DRIFT-Net: A Spectral-Coupled Neural Operator for PDEs Learning
Jiayi Li, Flora D. Salim
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 low-frequency mixing (ablation) | DRIFT-Net · NS-PwC | final-time rel L1: 1.65 -> 1.09 | Improved |
| add low-frequency mixing (ablation) | DRIFT-Net · NS-Tracer-PwC | final-time rel L1: 6.32 -> 4.19 | Improved |
| add radial gating (ablation) | DRIFT-Net · NS-PwC | final-time rel L1: 1.70 -> 1.09 | Improved |
| add radial gating (ablation) | DRIFT-Net · NS-Tracer-PwC | final-time rel L1: 6.55 -> 4.19 | Improved |
| add frequency-weighted loss (ablation) | DRIFT-Net · NS-PwC | final-time rel L1: 1.36 -> 1.09 | Improved |
| add frequency-weighted loss (ablation) | DRIFT-Net · NS-Tracer-PwC | final-time rel L1: 5.36 -> 4.19 | Improved |
| add 3x3 depth-wise conv (ablation) | DRIFT-Net · NS-PwC | final-time rel L1: 1.11 -> 1.09 | Improved |
| add 3x3 depth-wise conv (ablation) | DRIFT-Net · NS-Tracer-PwC | final-time rel L1: 4.36 -> 4.19 | Improved |
| add 1x1 pointwise linear (ablation) | DRIFT-Net · NS-PwC | final-time rel L1: 1.31 -> 1.09 | Improved |
| add 1x1 pointwise linear (ablation) | DRIFT-Net · NS-Tracer-PwC | final-time rel L1: 4.21 -> 4.19 | No clear effect |
| replace scOT windowed-attention blocks with DRIFT blocks | scOT · NS-PwC | final-time rel L1: 2.35 -> 1.09 | Improved |
| replace scOT windowed attention with DRIFT blocks | scOT · FNS-KF | final-time rel L1: 4.65 -> 4.32 | Improved |
Closest papers by meaning
- P$^2$C$^2$Net: PDE-Preserved Coarse Correction Network for efficient prediction of spatiotemporal dynamics (NeurIPS 2024 · similarity 0.90)
- Stable Long-Horizon PDE Forecasting via Latent Structured Spectral Propagators (NeurIPS 2026 · similarity 0.89)
- MultiPDENet: PDE-embedded Learning with Multi-time-stepping for Accelerated Flow Simulation (ICML 2025 · similarity 0.89)
- Spectral-Refiner: Accurate Fine-Tuning of Spatiotemporal Fourier Neural Operator for Turbulent Flows (ICLR 2025 · similarity 0.88)
- Hierarchical Implicit Neural Emulators (NeurIPS 2025 · similarity 0.88)
- PhyMPGN: Physics-encoded Message Passing Graph Network for spatiotemporal PDE systems (ICLR 2025 · similarity 0.88)
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.
