Causal Inference from Statistical Data

The so-called kernel-based tests of independence are developed for automatic causal discovery between random variables from purely observational statistical data, i.e., without intervention. Beyond the independence relations, the complexity of conditional distriubtions is used as an additional inference principle of determining the causal ordering between variables. Experiments with simulated and real-world data show that the proposed methods surpass the state-of-the-art approaches.