Identifiability Analysis of Linear ODE Systems with Hidden Confounders

Authors: Yuanyuan Wang, Biwei Huang, Wei Huang, Xi Geng, Mingming Gong

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To validate our theoretical results, we perform a series of simulations, which support and substantiate our findings. 5 Simulations. To evaluate the validity of the identifiability conditions established in Section 3 and 4, we present the results of simulations.
Researcher Affiliation Academia Yuanyuan Wang The University of Melbourne yuanyuanw2@student.unimelb.edu.au Biwei Huang University of California, San Diego bih007@ucsd.edu Wei Huang The University of Melbourne wei.huang@unimelb.edu.au Xi Geng The University of Melbourne xi.geng@unimelb.edu.au Mingming Gong The University of Melbourne mingming.gong@unimelb.edu.au
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes We provide all experimental details in Section 5 and include the code in the supplemental material.
Open Datasets No Observations are simulated from the true ODE systems for each case, with n equally-spaced observations generated from the time interval [0, 1] for each trajectory, and we only keep the values of the observable variables x.
Dataset Splits No The paper describes generating 'n equally-spaced observations' from simulated ODE systems but does not explicitly define or refer to standard training, validation, or test dataset splits.
Hardware Specification No Since our experiments solely consist of simulations designed to validate our theoretical findings, the computational resources employed are not a consideration for our research objectives.
Software Dependencies No The 'least_squares' function from the 'scipy.optimize' Python module, with default hyperparameter settings, is utilized for implementation.
Experiment Setup Yes The dimensions of both observable variables, d, and latent variables, p, are set to 3. Parameter initialization is performed near the true values to promote convergence to the global minimum. Specifically, for the η-(un)identifiable cases, initial parameter values are set to the true parameters plus a random value drawn from a uniform distribution U( 0.1, 0.1) for each replication. For {ηi}p 1-(un)identifiable cases, initial parameter values are set to the true values plus a random value from U( 0.3, 0.3).