Estimation of Bounds on Potential Outcomes For Decision Making

Authors: Maggie Makar, Fredrik Johansson, John Guttag, David Sontag

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our algorithm on a semi-synthetic clinical dataset and a well-known causality benchmark. We show how it can guide treatment decisions, and that it achieves a better trade-off between bound violations and utility than baseline algorithms.
Researcher Affiliation Academia 1CSAIL, MIT 2Chalmers University of Technology. Correspondence to: Maggie Makar <mmakar@mit.edu>.
Pseudocode Yes The procedure is summarized in Algorithm 1 in the supplement.
Open Source Code Yes Our code is available at <github.com/mymakar/bpo.git>.
Open Datasets Yes We use data from a randomized control trial measuring the effects of Heparin (International Stroke Trial Collaborative Group, 1997).
Dataset Splits Yes In each simulation, we randomly sample 3000 patients for training and validation and 3000 held out for testing.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts) used for running its experiments.
Software Dependencies No The paper mentions software components like 'RBF kernel', 'kernel regression', 'Gaussian processes', and 'logistic regression', but does not provide specific version numbers for any of these software dependencies.
Experiment Setup Yes We pick the regularization parameter for the propensity score model and all the response surface models via 3-fold cross-validation as described in detail in the supplement.