CUR Dynamic Mode Decomposition

Dynamic mode decomposition (DMD) extracts the dominant spatiotemporal modes from high-dimensional time-series data — the coherent structures that govern how a system evolves over time. It is widely used for fluid flows, video, and other measurement data where the dynamics are approximately linear in a suitable basis.

Standard DMD relies on the singular value decomposition, which becomes expensive as data grows. This project replaces the SVD step with a CUR approximation, which builds a low-rank factorization from a well-chosen subset of actual rows and columns of the data matrix. That reduces the computational cost on large-scale data, and keeps the factors interpretable — they are made of real measurements rather than abstract linear combinations.

The repository contains a NumPy implementation with examples of extracting and reconstructing dynamic modes from data. See the source code on GitHub.