A novel hypothesis-testing procedure that can effectively identify causal relationships between continuous variables, without any parametric assumptions, by leveraging proxy variables to adjust for the bias caused by unobserved confounders.
Integrating causal knowledge into local model-agnostic explanations to enhance transparency and reliability of black-box model decisions.
Effiziente Causal Discovery in stochastischen Prozessen durch Signature Kernel Tests.
Studying the effectiveness of a kernel-based conditional independence test in causal discovery for stochastic processes.
The author introduces a novel method, GIT, that leverages gradient estimators for targeted intervention in causal discovery.
The author develops a kernel-based test of conditional independence on path-space using signature kernels, demonstrating superior performance compared to existing approaches. The approach enables constraint-based causal discovery in acyclic stochastic dynamical systems.