Confounding-Robust Policy Improvement

Authors: Nathan Kallus, Angela Zhou

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We assess our methods on synthetic data and a large clinical trial, demonstrating that confounded selection can hinder policy learning and lead to unwarranted harm, while our robust approach guarantees safety and focuses on well-evidenced improvement.
Researcher Affiliation Academia Nathan Kallus Cornell University and Cornell Tech New York, NY kallus@cornell.edu Angela Zhou Cornell University and Cornell Tech New York, NY az434@cornell.edu
Pseudocode Yes Algorithm 1: Parametric Subgradient Method
Open Source Code No The paper does not contain any explicit statements about releasing source code or provide links to a code repository.
Open Datasets Yes We study the International Stroke Trial (IST)... [8] I. S. T. C. Group. The international stroke trial (ist): a randomised trial of aspirin, subcutaneous heparin, both, or neither among 19435 patients with acute ischaemic stroke. international stroke trial collaborative group. Lancet, 1997.
Dataset Splits No We construct an evaluation framework from the dataset by first sampling a split into a training set Strain and a held-out test set Stest, and subsampling a final set of initial patients, whose data is then used to train treatment assignment policies.
Hardware Specification No The paper does not specify any particular hardware used for conducting the experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions methods like 'causal forests' and 'logistic regression' but does not list any specific software packages or their version numbers that were used for implementation or analysis.
Experiment Setup No The paper states 'For each of these we vary the parameter Γ in {0.3, 0.4, . . . , 1.6, 1.7, 2, 3, 4, 5}' and 'For the parametric policies, we optimize with the same parameters as earlier' referring to Algorithm 1's step size and step-schedule exponent. However, specific values for hyperparameters like learning rate, batch size, or optimizer settings are not provided.