Bayesian Dynamic Mode Decomposition with Variational Matrix Factorization

Authors: Takahiro Kawashima, Hayaru Shouno, Hideitsu Hino8083-8091

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, both of nonlinear simulated and real-world datasets are used to illustrate the potential of the proposed method. In this section, we report the application of the proposed BDMD-VMF method to both simulated and real data. By using simulated data generated from nonlinear partial differential equations (PDEs), we confirm that BDMD-VMF yields almost the same results as DMD.
Researcher Affiliation Academia 1 The University of Electro-Communications, Tokyo, Japan 2 The Institute of Statistical Mathematics, Tokyo, Japan
Pseudocode Yes The DMD algorithm is as follows: 1. Compute the K-th truncated SVD of Y 0 Y 0 U KLKV K, where U K CD K and V K CT 1 K are composed of left and right singular vectors, respectively. The matrix LK CK K is a diagonal matrix consisting of the K largest singular values. Now, A can be approximated by A Y 1V KL 1 K U K. 2. Map A CD D to A CK K using the K left singular vectors U K of Y 0: A = U K AU K = U KY 1V KL 1 K CK K. 3. Compute the eigendecomposition of A A wk = λk wk, k = 1, . . . , K. This yields the K largest eigenvalues of A as {λk} and the corresponding eigenvectors, which are called DMD modes, as {U K wk}.
Open Source Code No The paper does not provide any concrete access to source code for the methodology described, such as a specific repository link, an explicit code release statement, or code in supplementary materials.
Open Datasets Yes As synthetic datasets, we used sequential datasets from two nonlinear PDEs: the nonlinear Schr odinger equation (NLSE) and Burgers equation. The Hu Ga DB dataset (Chereshnev and Kert esz-Farkas 2018) contains signals collected from six sensors placed on the right and left thighs, shins, and feet of 18 participants.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment.
Experiment Setup Yes Throughout the simulations reported below, we generated 7,500 MCMC samples and discarded the first 5,000 considering a burn-in periods. Unless otherwise specified, we employed appropriate weakly informative priors for any parameters. We determined the number of modes K for which DMD can capture input dynamics, through preliminary experiments.