ExDBN: Learning Dynamic Bayesian Networks using Extended Mixed-Integer Programming Formulations

Authors: Pavel Rytíř, Aleš Wodecki, Georgios Korpas, Jakub Marecek

TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comparing the novel approach to the state-of-the-art, we show that the proposed approach turns out to produce more accurate results when applied to small and medium-sized synthetic instances containing up to 80 time series. Lastly, two interesting applications in bioscience and finance, to which the method is directly applied, further stress the importance of developing highly accurate, globally convergent solvers that can handle instances of modest size. In Section 4, we conduct "Numerical Experiments" comparing Ex DBN with other methods like DYNOTEARS, Li NGAM, and NTS-NOTEARS, using synthetic data and real-world applications with metrics like SHD and F1 score.
Researcher Affiliation Collaboration Pavel Rytíř EMAIL Czech Technical University in Prague Aleš Wodecki EMAIL Czech Technical University in Prague Georgios Korpas EMAIL HSBC Holdings Plc., Singapore Czech Technical University in Prague Archimedes AI, Athena Research Center, Greece Jakub Mareček EMAIL Czech Technical University in Prague
Pseudocode No The paper describes the
Open Source Code No The text states: "The formulation and its implementation are easily reproducible, making it accessible to a wide range of potential practitioners." However, it does not provide a direct link to source code, nor an explicit statement that the code is being released or is available in supplementary materials. It only mentions general reproducibility.
Open Datasets No The paper mentions generating synthetic data in Section 4.1 but does not provide access to this generated data. For the finance application, it states: "the data from (Ballester et al., 2023) are not available from the authors, but we have downloaded 16 time-series capturing the spreads of 16 European CDS with RED6 codes..." This indicates they accessed data from a commercial source (implied Refinitiv), not a publicly available or open dataset.
Dataset Splits No The paper describes how synthetic data was generated in Section 4.1 and mentions
Hardware Specification No The paper mentions: "The running time for Ex DBN was capped at 7200 seconds, and the memory was limited to 32 GB." and later "the memory limit was increased to 128 GB." However, it does not specify any particular CPU or GPU models used for the experiments.
Software Dependencies Yes We used Gurobi as MIQP solver. Gurobi Optimization, LLC. Gurobi optimizer reference manual, 2024.
Experiment Setup No The paper mentions using regularization coefficients "λ, η > 0 are sufficiently small" and states "We apply a small threshold ϵ = 0.15 to West and At,est." While epsilon is a specific value, the regularization coefficients are only described qualitatively or with guidelines, not specific numeric values. For example, it states, "The regularization applied in Ex DBN needs to be scaled appropriately with the number of samples, as the optimal regularization constant is assumed to be a decreasing function of sample size." No concrete hyperparameter values or training configurations are provided in detail.