Research map / Probabilistic and scientific modeling / Differential equations and PDEs
Opening the Blackbox: Accelerating Neural Differential Equations by Regularizing Internal Solver Heuristics
Avik Pal, Yingbo Ma, Viral Shah, Christopher V Rackauckas
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 local-error regularization | Neural ODE · MNIST | prediction time: 0.094 $\pm$ 0.010 -> 0.060 $\pm$ 0.001 | Improved |
| add stiffness regularization (ablation) | Neural ODE · MNIST | prediction time: 0.094 $\pm$ 0.010 -> 0.094 $\pm$ 0.003 | No clear effect |
| combine stiffness and local-error regularization (ablation) | Neural ODE · MNIST | prediction time: 0.094 $\pm$ 0.010 -> 0.081 $\pm$ 0.006 | Improved |
| add local-error regularization | Latent ODE · PhysioNet 2012 | training time: 1.75 $\pm$ 0.39 -> 0.94 $\pm$ 0.13 | Improved |
| add stiffness regularization (ablation) | Latent ODE · PhysioNet 2012 | prediction time: 0.53 $\pm$ 0.12 -> 0.20 $\pm$ 0.01 | Improved |
| replace the local-error penalty with squared error estimates (ablation) | Latent ODE · PhysioNet 2012 | test results | No clear effect |
| add local-error regularization | Neural SDE · Spiral SDE | training time: 178.95 $\pm$ 20.22 -> 173.43 $\pm$ 04.18 | Improved |
| add stiffness regularization (ablation) | Neural SDE · Spiral SDE | prediction time: 0.07553 $\pm$ 0.0186 -> 0.07250 $\pm$ 0.0017 | Improved |
| add local-error regularization | Neural SDE · MNIST | prediction time: 15.07 $\pm$ 0.93 -> 07.23 $\pm$ 0.14 | Improved |
| add stiffness regularization (ablation) | Neural SDE · MNIST | training time: 6.32 $\pm$ 0.19 -> 8.54 $\pm$ 0.37 | Hurt |
| use higher-order derivative regularization | Neural ODE · PhysioNet 2012 | training time: 1.75 $\pm$ 0.39 -> 12.3 $\pm$ 0.32 | Hurt |
Closest papers by meaning
- Locally Regularized Neural Differential Equations: Some Black Boxes were meant to remain closed! (ICML 2023 · similarity 0.95)
- Neural Ordinary Differential Equations (NeurIPS 2018 · similarity 0.90)
- Learning Differential Equations that are Easy to Solve (NeurIPS 2020 · similarity 0.90)
- Breakeven complexity: A new perspective on neural partial differential equation solvers (NeurIPS 2026 · similarity 0.89)
- Efficient Training of Neural Fractional-Order Differential Equation via Adjoint Backpropagation (AAAI 2025 · similarity 0.88)
- On the Generalization and Approximation Capacities of Neural Controlled Differential Equations (ICLR 2024 · 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.
