Adversarial Causal Bayesian Optimization

Authors: Scott Sussex, Pier Giuseppe Sessa, Anastasia Makarova, Andreas Krause

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, CBO-MW outperforms non-causal and non-adversarial Bayesian optimization methods on synthetic environments and environments based on real-word data.
Researcher Affiliation Academia Scott Sussex ETH Zürich scott.sussex@inf.ethz.ch Pier Giuseppe Sessa ETH Zürich Anastasiia Makarova ETH Zürich Andreas Krause ETH Zürich
Pseudocode Yes Algorithm 1 Causal Bayesian Optimization Multiplicative Weights (CBO-MW), Algorithm 2 Causal UCB Oracle, Algorithm 3 Multiplicative Weights Update (MWU), Algorithm 4 Distributed Causal Bayesian Optimization Multiplicative Weights (D-CBO-MW)
Open Source Code Yes Attached to this submission we include code (https://github.com/ssethz/acbo) that implements CBO-MW, D-CBO-MW and all baselines seen in the experiments.
Open Datasets Yes A simulator is constructed using historical data from an SMS in Louisville, KY (Lou, 2021). The demand for trips corresponds to real trip data from Lou (2021). Louisville Kentucky Open Data, 2021. URL https://data.louisvilleky.gov/ dataset/dockless-vehicles.
Dataset Splits No The paper evaluates methods over 10 repeats and uses the first 10 days for model initialization, but does not specify explicit training/validation/test dataset splits needed for reproducibility (e.g., percentages or sample counts for each split, or k-fold cross-validation setup).
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or cloud instance types) used for running its experiments.
Software Dependencies No The paper provides a link to source code but does not explicitly list specific software dependencies with version numbers (e.g., Python 3.x, PyTorch x.y) in the text.
Experiment Setup No The paper mentions performing a hyperparameter sweep to select the best hyperparameters and notes initialization details (e.g., '2m + 1 samples'), but it does not explicitly list the specific values of hyperparameters (e.g., learning rates, batch sizes, specific kernel function parameters, or the values for τ and βt) used in the experiments.