Bounds on Representation-Induced Confounding Bias for Treatment Effect Estimation

Authors: Valentyn Melnychuk, Dennis Frauen, Stefan Feuerriegel

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our bounds in a series of experiments.
Researcher Affiliation Academia Valentyn Melnychuk, Dennis Frauen & Stefan Feuerriegel LMU Munich & Munich Center for Machine Learning Munich, Germany
Pseudocode No The paper describes its neural refutation framework in three stages (Stage 0, Stage 1, Stage 2) but does not provide pseudocode or a formally labeled algorithm block.
Open Source Code Yes 1Code is available at https://github.com/Valentyn1997/RICB.
Open Datasets Yes IHDP100 dataset. The Infant Health and Development Program (IHDP) (Hill, 2011; Shalit et al., 2017) is a classical benchmark for CATE estimation... HC-MNIST dataset. HC-MNIST is a semi-synthetic benchmark on top of the MNIST image dataset (Jesson et al., 2021). The MNIST dataset contains ntrain = 60, 000 train and ntest = 10, 000 test images. (Le Cun, 1998)
Dataset Splits Yes We performed hyperparameter tuning at all the stages of our refutation framework for all the networks based on five-fold cross-validation using the training subset.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., CPU, GPU models, memory, or cluster specifications) used for running its experiments.
Software Dependencies No The paper mentions software like 'Py Torch and Pyro' and optimizers like 'Adam W', but does not provide specific version numbers for these software dependencies.
Experiment Setup Yes Implementation. ... Each network was trained with niter = 5, 000 train iterations. Hyperparameters. We performed hyperparameter tuning at all the stages of our refutation framework for all the networks based on five-fold cross-validation using the training subset. At each stage, we did a random grid search with respect to different tuning criteria. Table 5 provides all the details on hyperparameters tuning.