Scalable Variational Inference for Dynamical Systems

Authors: Nico S. Gorbach, Stefan Bauer, Joachim M. Buhmann

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In order to provide a fair comparison to existing approaches, we test our approach on two small to medium sized ODE models, which have been extensively studied in the same parameter settings before [e.g. Calderhead et al., 2008, Dondelinger et al., 2013, Wang and Barber, 2014]. Additionally, we show the scalability of our approach on a large-scale partially observable system which has so far been infeasible to analyze with existing gradient matching methods due to the number of unobserved states.
Researcher Affiliation Academia Nico S. Gorbach Dept. of Computer Science ETH Zurich ngorbach@inf.ethz.ch Stefan Bauer Dept. of Computer Science ETH Zurich bauers@inf.ethz.ch Joachim M. Buhmann Dept. of Computer Science ETH Zurich jbuhmann@inf.ethz.ch
Pseudocode Yes Algorithm 1 Mean-field coordinate ascent for GP Gradient Matching
Open Source Code No The paper does not provide an explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets No The data used for experiments is simulated, not from a publicly available dataset. For example: 'We used the same ODE parameters as in Dondelinger et al. [2013]...to simulate the data...Mean-field variational inference for gradient matching was performed on a simulated dataset with additive Gaussian noise with variance σ2 = 0.25.'
Dataset Splits No The paper does not specify training, validation, or test dataset splits. The data used in the experiments is simulated based on given parameters or time points, not partitioned from a larger corpus.
Hardware Specification Yes All experiments were run on a 2.5 GHz Intel Core i7 Macbook.
Software Dependencies No The paper does not provide specific software names with version numbers for reproducibility (e.g., 'Python 3.8', 'PyTorch 1.9').
Experiment Setup Yes We used the same ODE parameters as in Dondelinger et al. [2013] (i.e. θ1 = 2, θ2 = 1, θ3 = 4, θ4 = 1) to simulate the data over an interval [0, 2] with a sampling interval of 0.1. Predator species (i.e. x1) were initialized to 3 and prey species (i.e. x) were initialized to 5. Mean-field variational inference for gradient matching was performed on a simulated dataset with additive Gaussian noise with variance σ2 = 0.25.