Research map / Probabilistic and scientific modeling / Dynamical systems
Learning Chaotic Dynamics in Dissipative Systems
Zongyi Li, Miguel Liu-Schiaffini, Nikola Borislavov Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, Anima Anandkumar
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 dissipativity regularization to the learned operator (ablation) | Markov neural operator (MNO) · Kolmogorov flow Re=500 | stability under perturbed initial conditions | Improved |
| add dissipativity regularization (ablation) | MNO · Kolmogorov flow Re=5000 | accuracy | No clear effect |
| vary dissipativity hyperparameters (loss weight, scaling factor, shell radius) (ablation) | MNO · Lorenz-63 | step-wise and dissipative error | No clear effect |
| learn identity residual instead of solution operator at small time step h (ablation) | MNO · Kuramoto-Sivashinsky | per-step and accumulated error | Improved |
| learn identity residual instead of solution operator at large time step h (ablation) | MNO · Kuramoto-Sivashinsky | error | No clear effect |
| MNO instead of LSTM and GRU | evolution operator model · Kuramoto-Sivashinsky h=1s | Fourier spectrum accuracy | Improved |
| MNO instead of LSTM and GRU | evolution operator model · Kuramoto-Sivashinsky h=1s | other invariant statistics | No clear effect |
| MNO instead of U-Net and LSTM-CNN | vorticity predictor · Navier-Stokes Re=40 | vorticity error | Improved |
| train with H2 loss instead of L2 or H1 (ablation) | MNO · Navier-Stokes Re=40 | vorticity error in L2, H1, H2 norms | Improved |
| train with L2 or H1 loss instead of H2 (ablation) | MNO · Navier-Stokes Re=40 | TKE and dissipation error | Mixed |
| use derivative order slightly higher than the target quantity (e.g. velocity plus H1 loss for velocity) (ablation) | MNO · Kolmogorov flow Re=500 | velocity error | Improved |
| use much higher derivative order loss (ablation) | MNO · Kolmogorov flow | invariant statistics vs resolution | Mixed |
Closest papers by meaning
- Chaos Meets Attention: Transformers for Large-Scale Dynamical Prediction (ICML 2025 · similarity 0.91)
- DySLIM: Dynamics Stable Learning by Invariant Measure for Chaotic Systems (ICML 2024 · similarity 0.90)
- Learning invariant representations of time-homogeneous stochastic dynamical systems (ICLR 2024 · similarity 0.87)
- When are dynamical systems learned from time series data statistically accurate? (NeurIPS 2024 · similarity 0.87)
- Learning Deep Dissipative Dynamics (AAAI 2025 · similarity 0.86)
- The Dynamic-Probabilistic Consistency Gap in Chaotic Surrogate Modeling (NeurIPS 2026 · similarity 0.86)
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.
