Constrained Causal Bayesian Optimization

Authors: Virginia Aglietti, Alan Malek, Ira Ktena, Silvia Chiappa

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate c CBO on artificial and real-world causal graphs showing successful trade off between fast convergence and percentage of feasible interventions.
Researcher Affiliation Industry 1Deep Mind, London, UK.
Pseudocode Yes Algorithm 1 c CBO
Open Source Code Yes Code for reproducing the experiments is available at https://github.com/deepmind/ccbo.
Open Datasets Yes For the HEALTH graph, we use the SCM in Ferro et al. (2015)... For the PROTEIN-SIGNALING graph, we use the observational dataset from Sachs et al. (2005)
Dataset Splits No The paper does not explicitly mention train/validation/test splits, nor does it provide percentages or sample counts for such splits. It discusses observational and interventional datasets used during sequential experimentation.
Hardware Specification No No specific hardware details (e.g., CPU, GPU models, memory) are provided for the experimental setup.
Software Dependencies No No specific software dependencies with version numbers are mentioned in the paper.
Experiment Setup Yes For all experiments we initialize the kernel hyper-parameters (σ2 f, l) of the surrogate models to 1 and optimize them with a standard type-2 ML approach. When sampling from the modified SCM is required in order to compute the prior mean or kernel functions we use S = 10 samples. The same number of samples is used when the computation of the acquisition function requires a Monte Carlo approximation.