Research map / Probabilistic and scientific modeling / Dynamical systems
Compositional Modeling of Nonlinear Dynamical Systems with ODE-based Random Features
Thomas M. McDonald, Mauricio A Álvarez
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 |
|---|---|---|---|
| Replace EQ random features with ODE-based random features in the deep model | DLFM versus DGP-EQ · CHARIS extrapolation, ABP and ECG | NMSE: 0.73 -> 0.39; 0.66 -> 0.45 | Improved |
| Replace EQ random features with ODE-based random features in the deep model | DLFM versus DGP-EQ · CHARIS extrapolation, ABP and ECG | MNLL: 1.21 -> 0.75; 1.22 -> 0.99 | Improved |
| Use two latent forces instead of one (ablation) | DLFM · CHARIS extrapolation, ABP, ECG, and ICP | NMSE: ABP: 0.39 -> 0.42; ECG: 0.45 -> 1.58; ICP: 0.36 -> 0.32 | Mixed |
| Use one latent force instead of two (ablation) | DLFM · CHARIS extrapolation, ABP and ECG | MNLL: ABP: 0.77 -> 0.75; ECG: 1.16 -> 0.99 | Improved |
| Use two latent forces instead of one (ablation) | DLFM · CHARIS extrapolation, ICP | MNLL: 0.90 -> 0.88 | Improved |
| Use ODE-based random features rather than EQ random features | DLFM versus DGP-EQ · Powerplant test | NMSE: 0.0581 -> 0.0557 | Improved |
| Use ODE-based random features rather than EQ random features | DLFM versus DGP-EQ · Powerplant test | MNLL: 0.016 -> - 0.017 | Improved |
| Use ODE-based random features rather than EQ random features | DLFM versus DGP-EQ · Protein test | MNLL: 1.101 -> 0.754 | Improved |
| Use the DLFM rather than DGP-EQ | DLFM · Protein test | NMSE: 0.553 -> 0.698 | Hurt |
| Use the DLFM rather than DGP-EQ | DLFM · Toy system extrapolation | MNLL: 3.7 -> 2.4 | Improved |
| Use the DLFM rather than C-DGP | DLFM · FitzHugh-Nagumo, full-data fit | NMSE: 0.01 -> 0.02 | Hurt |
| Use the DLFM rather than C-DGP | DLFM · FitzHugh-Nagumo extrapolation | NMSE: 0.38 -> 0.07 | Improved |
Closest papers by meaning
- Streaming Variational Monte Carlo (TPAMI 2022 · similarity 0.88)
- Overcoming Mean-Field Approximations in Recurrent Gaussian Process Models (ICML 2019 · similarity 0.88)
- Closing the Approximation Gap in Simulation-free Latent SDEs (NeurIPS 2026 · similarity 0.87)
- SING: SDE Inference via Natural Gradients (NeurIPS 2025 · similarity 0.87)
- Predicting partially observable dynamical systems via diffusion models with a multiscale inference scheme (NeurIPS 2025 · similarity 0.87)
- Variational Inference for SDEs Driven by Fractional Noise (ICLR 2024 · similarity 0.87)
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.
