Fast Approximate Dynamic Programming for Infinite-Horizon Markov Decision Processes

Authors: Mohamad Amin Sharifi Kolarijani, Gyula Max, Peyman Mohajerin Mohajerin Esfahani

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

Reproducibility Variable Result LLM Response
Research Type Experimental 4 Numerical simulations
Researcher Affiliation Academia M. A. S. Kolarijani Delft Center for Systems and Control Delft University of Technology The Netherlands M.A.Sharifi Kolarijani@tudelft.nl G. F. Max Delft Center for Systems and Control Delft University of Technology The Netherlands G.F.Max@tudelft.nl P. Mohajerin Esfahani Delft Center for Systems and Control Delft University of Technology The Netherlands P.Mohajerin Esfahani@tudelft.nl
Pseudocode Yes Algorithm 1 Conj VI: Approximate VI in conjugate domain
Open Source Code Yes We also provide the Conj VI MATLAB package [21] for the implementation of the proposed algorithm. The package also includes the numerical simulations of this section. and the reference '[21] M. A. S. Kolarijani and P. Mohajerin Esfahani. Conjugate value iteration (Conj VI) MATLAB package. Licensed under the MIT License, available online at https://github.com/Amin Kolarijani/ Conj VI, 2021.'
Open Datasets No The paper uses simulated control problems ('noisy inverted pendulum' and 'unstable deterministic batch reactor') based on setups borrowed from cited works [19, 20]. It does not involve using or providing access to publicly available pre-collected datasets in the conventional sense for training or evaluation.
Dataset Splits No The paper describes optimal control problems solved via simulation and does not refer to dataset splits for training, validation, or testing.
Hardware Specification No The paper does not provide specific hardware details such as CPU/GPU models, memory, or specific computing environments used for running the experiments.
Software Dependencies No The paper mentions the use of 'MATLAB package' and 'LLTd routine' but does not specify version numbers for these software dependencies, which would be necessary for reproducible replication.
Experiment Setup Yes The paper specifies experimental setup details such as the use of 'nearest neighbor extension' or 'multi-linear interpolation and extrapolation for the extension operators', 'dynamic scheme for the construction of Yg', and that 'all of the involved discrete sets are uniform grids with the same size N in each dimension', with specific values for N like 'N = 41' and 'N = 25'.