2204.02373
Data-driven Influence Based Clustering of Dynamical Systems
Subhrajit Sinha
correctmedium confidence
- Category
- Not specified
- Journal tier
- Specialist/Solid
- Processed
- Sep 28, 2025, 12:56 AM
- arXiv Links
- Abstract ↗PDF ↗
Audit review
The paper states the one-step information transfer formula for linear Gaussian systems and a data-driven Koopman/DMD procedure; the model explicitly derives the same closed-form via conditional-Gaussian/Schur-complement and law-of-total-variance, and details a compatible influence-graph construction. Minor procedural differences exist in the data-driven implementation (noise term λ vs σ2 and whether to re-propagate covariance under the frozen model), but these do not contradict the core results.
Referee report (LaTeX)
\textbf{Recommendation:} minor revisions
\textbf{Journal Tier:} specialist/solid
\textbf{Justification:}
The submission connects an information-theoretic causality measure to a practical influence distance and clustering pipeline, and demonstrates utility on multiple systems. The main linear-Gaussian expression and freeze semantics are correct, and the Koopman/DMD integration is appropriate. However, the use of λ inside the entropy term and the covariance handling under freezing should be clarified and contrasted with the analytic one-step definition. These are minor but important points for rigor and reproducibility.