The arXiv abstract introduces DeDiCa, a framework for efficient causal inference when working with discrete variables and hidden confounding. The authors say it revisits classical probabilistic modeling through a tensorized parameterization of causal Bayesian networks, which appears to be the core technical contribution.
Because the abstract is truncated, the announcement does not specify the exact algorithms, experiments, or comparisons. What is clear is that the work is positioned as practical and scalable, suggesting the tensorized approach is meant to address computational bottlenecks in existing deconfounding methods.
Readers interested in the full methodology will need to consult the paper itself, as the abstract only sketches the motivation and the proposed parameterization.