Bayesian Active Causal Discovery with Multi-Fidelity Experiments

Authors: Zeyu Zhang, Chaozhuo Li, Xu Chen, Xing Xie

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments to demonstrate the effectiveness of our model. In this section, we conduct experiments to demonstrate the effectiveness of our model, where we focus on the following problems: (1) whether our model can achieve better performance than the previous ACD methods? (2) Whether the constraint in objective (10) in necessary? (3) How the DAG regularization coefficient influence the model performance? In the following, we first introduce the experiment setup, and then present and analyze the results.
Researcher Affiliation Collaboration Zeyu Zhang Gaoling School of Artificial Intelligence Renmin University of China zeyuzhang@ruc.edu.cn Chaozhuo Li Microsoft Research Asia cli@microsoft.com Xu Chen Gaoling School of Artificial Intelligence Renmin University of China xu.chen@ruc.edu.cn Xing Xie Microsoft Research Asia xingx@microsoft.com
Pseudocode Yes Algorithm 1: Algorithm of Licence for Single Intervention Scenario. Algorithm 2: Algorithm of Licence for Batch Intervention Scenario.
Open Source Code No The paper mentions 'the configuration file in our codes' but does not provide an explicit statement of code release or a link to a public repository for the methodology described.
Open Datasets Yes We experiment with three commonly used causal discovery datasets, including Erd os-Rényi graph (ER) [41], Scale-Free graph (SF) [42] and DREAM [43].
Dataset Splits No The paper does not explicitly provide specific training/test/validation dataset splits. It mentions initial observational samples and then acquiring interventional data, but without precise percentages, counts, or predefined split references for reproduction.
Hardware Specification Yes Name Details CPU Intel Xeon Platinum 8350C 2.60GHz GPU RTX A5000 (24GB) Memory 42GB RAM
Software Dependencies Yes Name Details Python Version 3.8 Java Version 1.8.0 (Necessary for DREAM)
Experiment Setup Yes Table 2: The details of experimental settings. Name Explanation Value budget The total budget for interventional experiments, (i.e., C). 10/20/30/40/50 oracle number The number of oracles, (i.e., M) 3 oracle cost The cost for each oracle, (i.e., Λ) 2, 8, 32 oracle noise The extra additive noise for each oracle. 0.04, 0.02, 0.00 observation number The number of observational samples. 1000 expect edge number The number of expect edges. 2 additive noise The value of additive noise during data generations. 0.01. We carefully tune the hyper-parameters for baselines and our model, and the final values can be obtained in the configuration file in our codes.