Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Learning Functional Transduction
Mathieu Chalvidal, Thomas Serre, Rufin VanRullen
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 meta-trained Transducer instead of per-instance FNO | Transducer · ADR PDE regression | RMSE: 2.96e^{-4} -> 2.39e^{-4} | Improved |
| use Transducer instead of DeepONet | Transducer · ADR PDE regression | RMSE: 2.02e^{-2} -> 2.39e^{-4} | Improved |
| single feedforward pass of pre-trained Transducer instead of fitting FNO from scratch | Transducer · ADR PDE regression | regression time (s): 1.72e^{2} -> 3.10e^{-3} | Improved |
| Transducer instead of FNO for compute | Transducer · ADR PDE regression | GFLOPs: 1.68e^{2} -> 1.06e^{-1} | Improved |
| increase number of examples beyond 60 | FNO vs Transducer · ADR PDE regression | RMSE vs dataset size | Mixed |
| increase Transducer depth with untied weights (ablation) | Transducer · ADR PDE regression | training loss | Improved |
| replace exp dot product kernel with RBF kernel (ablation) | Transducer · ADR PDE regression, s=10 | RMSE: 2.71e-3 -> 8.71e-3 | Hurt |
| replace exp dot product kernel with RBF kernel (ablation) | Transducer · ADR PDE regression, s=500 | RMSE: 1.79e-4 -> 3.22e-4 | Hurt |
| replace exp dot product kernel with l2-norm kernel (ablation) | Transducer · ADR PDE regression, s=100 | RMSE: 2.39e-4 -> 6.98e-4 | Hurt |
| replace exp dot product kernel with l2-norm kernel, test at s=500 beyond meta-training sizes (ablation) | Transducer · ADR PDE regression, s=500 | RMSE: 1.79e-4 -> 7.33e5 | Hurt |
| condition on previous-year window (P.Y) instead of ViT baseline | Transducer · ERA5 pressure | LWMSE (hPa): 32.826 -> 25.293 | Improved |
| condition on 15-day window lagging one week (P.W) instead of previous year window (P.Y) | Transducer · ERA5 pressure | LWMSE (hPa): 25.293 -> 22.718 | Improved |
Closest papers by meaning
- On Hypothesis Transfer Learning of Functional Linear Models (ICML 2024 · similarity 0.85)
- Optimal deep learning of holomorphic operators between Banach spaces (NeurIPS 2024 · similarity 0.84)
- NOFE – Neural Operator Function Embedding (NeurIPS 2026 · similarity 0.83)
- Improving Adaptivity via Over-Parameterization in Sequence Models (NeurIPS 2024 · similarity 0.83)
- Thompson Sampling in Function Spaces via Neural Operators (NeurIPS 2025 · similarity 0.83)
- Neural Operators with Localized Integral and Differential Kernels (ICML 2024 · similarity 0.83)
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.
