A Neural Framework for Generalized Causal Sensitivity Analysis

Authors: Dennis Frauen, Fergus Imrie, Alicia Curth, Valentyn Melnychuk, Stefan Feuerriegel, Mihaela van der Schaar

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

Reproducibility Variable Result LLM Response
Research Type Experimental We provide theoretical guarantees that NEURALCSA can infer valid bounds on the causal query of interest and also demonstrate this empirically using both simulated and real-world data.
Researcher Affiliation Academia 1 LMU Munich 2 Munich Center for Machine Learning 3 UCLA 4 University of Cambridge 5 Alan Turing Institute
Pseudocode Yes Algorithm 1: Full learning algorithm for NEURALCSA
Open Source Code Yes 7 Code is available at https://github.com/Dennis Frauen/Neural CSA.
Open Datasets Yes We create a semi-synthetic dataset using MIMIC-III (Johnson et al., 2016)...
Dataset Splits Yes We obtain n = 14719 patients and split the data into train (80%), val (10%), and test (10%).
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory) used for running experiments are mentioned in the paper.
Software Dependencies No The paper mentions software components like 'autoregressive neural spline flows' and 'Adam optimizer' but does not specify their version numbers or the version of the programming language used.
Experiment Setup Yes We choose the number of epochs such that NEURALCSA satisfies the sensitivity constraint for a given sensitivity parameter. Details are in Appendix F. ... Table 4: Hyperparameter tuning details. MODEL TUNABLE PARAMETERS SEARCH RANGE Stage 1 CNF Epochs 50 Batch size 32, 64, 128 Learning rate 0.0005, 0.001, 0.005 Hidden layer size (hyper network) 5, 10, 20, 30 Number of spline bins 2, 4, 8 Propensity network Epochs 30 Batch size 32, 64, 128 Learning rate 0.0005, 0.001, 0.005 Hidden layer size 5, 10, 20, 30 Dropout probability 0, 0.1