First-Order Methods for Wasserstein Distributionally Robust MDP

Authors: Julien Grand Clement, Christian Kroer

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

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments show that our algorithm is significantly more scalable than state-of-the-art approaches across several domains.
Researcher Affiliation Academia 1IEOR Department, Columbia University.
Pseudocode Yes Algorithm 1 First-order Method for Wasserstein DR-MDP
Open Source Code No The paper does not provide any specific links to source code or state that the code is publicly available.
Open Datasets No For each MDP instance, we generate the sampled kernels ˆy1, ..., ˆy N by considering N small random (Garnet) perturbations around the true nominal kernel y0 (see Appendix F).
Dataset Splits No The paper describes generating MDP instances and evaluating performance based on running times and optimality criteria, but it does not specify train/validation/test dataset splits for a fixed dataset.
Hardware Specification Yes We run our simulations on a laptop with 2.2 GHz Intel Core i7 and 8 GB of RAM.
Software Dependencies Yes We implement our algorithms in Python 3.7.3, using Gurobi 8.1.1 to solve any linear/quadratic optimization program involved.
Experiment Setup Yes All figures in this section show the running times of the algorithms before returning an ϵ-optimal policy with ϵ = 0.1. We initialize all algorithms with v0 = 0. The running times are averaged across 5 instances by changing the seeds for sampling the N kernels around y0.