Markov Chain Monte Carlo for Continuous-Time Switching Dynamical Systems

Authors: Lukas Köhs, Bastian Alt, Heinz Koeppl

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our framework under the modeling assumption and compare it against an existing variational inference approach. 4.1. Verification on Ground-Truth Data 1D System We test the method on synthetic data from a one-dimensional two-mode switching dynamical system as specified in Section 2.2. Our Gibbs sampling scheme is able to faithfully recover the latent ground-truth trajectories; both z[0,T ] and y[0,T ] are reproduced with high fidelity, see Fig. 2. 4.2. Inference of Gene-Switching Dynamics We use our framework to infer the switching dynamics of an inducible gene system measured in-house.
Researcher Affiliation Academia 1Department of Electrical Engineering and Information Technology, Technische Universität Darmstadt, Darmstadt, Germany. Correspondence to: Heinz Koeppl <heinz.koeppl@tudarmstadt.de>.
Pseudocode Yes Algorithm 1 BFFB Gibbs Sampling for Continuous-Time Switching Dynamical Systems
Open Source Code Yes An implementation of the proposed framework is publicly available. https://git.rwth-aachen.de/bcs/projects/lk/mcmc-ct-sds.git
Open Datasets No The paper mentions using "synthetic data" and an "inducible gene expression system measured in-house" (Figure 4). While data is used, there is no specific link, DOI, or citation for a publicly available dataset. The synthetic data is generated by the authors, and the gene expression data is in-house.
Dataset Splits No The paper mentions using "synthetic data" and "in-house" data for verification and inference, respectively, but does not explicitly provide training, validation, or test dataset splits. It evaluates on the full generated/collected data.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory specifications) used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., Python, PyTorch, or other libraries/solvers).
Experiment Setup Yes In all experiments, the hyperparameters are set empirically; for other options, see, e.g. (Casella, 2001). We initialize the Gibbs sampler in the same way to start at reasonable parameter values such as to achieve fast burn-in. All hyperparameters are provided in Appendix D. ... D.1. Hyperparameter Settings We initialize all distribution hyperparameters, cf. Appendix C, empirically.