2021/05/11 by Marios Papamichalis, Papamichalis, Marios, Abhishek Ray +7
Computer Science · Mathematics · #Advanced Causal Inference Techniques #Artificial Intelligence (cs.AI) #Bayesian Modeling and Causal Inference #FOS: Computer and information sciences #Statistical Methods and Inference
paper · pdf · doi:10.48550/arxiv.2105.05395
openalex publication_date 2021/05/11 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Probabilistic machine learning models are often insufficient to help with\ndecisions on interventions because those models find correlations - not causal\nrelationships. If observational data is only available and experimentation are\ninfeasible, the correct approach to study the impact of an intervention is to\ninvoke Pearl's causality framework. Even that framework assumes that the\nunderlying causal graph is known, which is seldom the case in practice. When\nthe causal structure is not known, one may use out-of-the-box algorithms to\nfind causal dependencies from observational data. However, there exists no\nmethod that also accounts for the decision-maker's prior knowledge when\ndeveloping the causal structure either. The objective of this paper is to\ndevelop rational approaches for making decisions from observational data in the\npresence of causal graph uncertainty and prior knowledge from the\ndecision-maker. We use ensemble methods like Bayesian Model Averaging (BMA) to\ninfer set of causal graphs that can represent the data generation process. We\nprovide decisions by computing the expected value and risk of potential\ninterventions explicitly. We demonstrate our approach by applying them in\ndifferent example contexts.\n