Research map / Probabilistic and scientific modeling / Dynamical systems
Learning Compositional Koopman Operators for Model-Based Control
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 relation-based shared blocks in the Koopman matrix instead of a diagonal-only structure (ablation) | Compositional Koopman Operators · Rope | simulation error: 0.133 -> 0.105; 0.174 -> 0.075 | Helped |
| use relation-based shared blocks in the Koopman matrix instead of a diagonal-only structure (ablation) | Compositional Koopman Operators · Rope | control error: 2.337 -> 0.854; 2.809 -> 1.101 | Helped |
| add relation-based shared blocks to an unstructured Koopman matrix (ablation) | Compositional Koopman Operators · Rope | simulation error: 0.117 -> 0.105; 0.083 -> 0.075 | Helped |
| add relation-based shared blocks to an unstructured Koopman matrix (ablation) | Compositional Koopman Operators · Rope | control error: 1.522 -> 0.854; 1.288 -> 1.101 | Helped |
| vary the number of data samples for system identification (ablation) | Compositional Koopman Operators · Rope | simulation error | Helped |
| vary the per-object Koopman embedding dimension (ablation) | Compositional Koopman Operators · Rope | simulation error | Mixed |
| add the metric loss to encourage distance-preserving embeddings (ablation) | Compositional Koopman Operators · Rope | control error | Helped |
| add the metric loss to encourage distance-preserving embeddings (ablation) | Compositional Koopman Operators · Rope | simulation error | No clear effect |
| identify dynamics by fitting transition matrices rather than optimizing physical parameters with Bayesian Optimization | Rope adaptation with unknown physical parameters · Rope | adaptation time: 0.43 vs. 180 seconds | Helped |
| use compositional Koopman identification instead of Bayesian Optimization with an approximate spring-mass model | Rope adaptation · Rope | simulation error: 0.046 vs. 0.084 | Helped |
| use compositional Koopman identification instead of Bayesian Optimization with an approximate spring-mass model | Rope adaptation · Rope | control error: 0.854 vs. 2.547 | Helped |
| provide ground-truth physical parameters to IN/PN instead of adapting when those parameters are unknown | Compositional Koopman Operators versus IN/PN · known physical parameter setting | performance | Mixed |
Closest papers by meaning
- Deep-Koopman-KANDy: Dictionary Discovery for Deep-Koopman Operators with Kolmogorov-Arnold Networks for Dynamics (NeurIPS 2026 · similarity 0.88)
- Optimizing Neural Networks via Koopman Operator Theory (NeurIPS 2020 · similarity 0.88)
- Efficient Parametric SVD of Koopman Operator for Stochastic Dynamical Systems (NeurIPS 2025 · similarity 0.86)
- ResKoopNet: Learning Koopman Representations for Complex Dynamics with Spectral Residuals (ICML 2025 · similarity 0.85)
- Conformal Online Learning of Deep Koopman Linear Embeddings (NeurIPS 2025 · similarity 0.84)
- Course Correcting Koopman Representations (ICLR 2024 · similarity 0.84)
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.