Off-Policy Interval Estimation with Lipschitz Value Iteration

Authors: Ziyang Tang, Yihao Feng, Na Zhang, Jian Peng, Qiang Liu

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We test our algorithm on a number of benchmarks and show that it can provide tight and provably correct bounds.
Researcher Affiliation Academia Ziyang Tang University of Texas at Austin ztang@utexas.edu Yihao Feng University of Texas at Austin yihao@cs.utexas.edu Na Zhang Tsinghua University zhangna@pbcsf.tsinghua.edu.cn Jian Peng University of Illinois at Urbana-Champaign jianpeng@illinois.edu Qiang Liu University of Texas at Austin lqiang@cs.utexas.edu
Pseudocode Yes Algorithm 1 Lipschitz Value Iteration (for Upper Bound); Algorithm 2 Lipschitz Value (Upper Bound) Iteration with Stochastic Update
Open Source Code No The paper does not provide concrete access to source code for the methodology described. There is no explicit statement of code release, nor a link to a code repository.
Open Datasets No The paper mentions using "Transition data D = {si, ai, s i, ri}1 i n" and performs experiments on "Synthesis Environment with A Known Value Function", "Pendulum Environment", and "HIV Simulator". While these are known environments, the paper does not provide specific access information (link, DOI, citation with authors/year) for a publicly available or open dataset that was used for training or was made available by the authors for replication.
Dataset Splits No The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) for a validation set.
Hardware Specification No The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes The default settings: number of trajectory nt = 30, Horizon length H = 100, discounted factor γ = 0.95, Lipschitz constant η = 2.0 and subsample size n B = 500. We run Lipschitz value iteration for 100 iteration to ensure almost convergence.