QACD: A Quantitative Argumentation Framework for Causal Discovery

A new framework called Quantitative Argumentation for Causal Discovery (QACD) has been introduced by researchers. This semantics-driven approach views outcomes of conditional independence as graded, defeasible arguments rather than as absolute constraints. It translates results from statistical tests into argument strengths and reconciles conflicting evidence through connectivity-mediated witness propagation, resulting in a fixed-point acceptability labeling for candidate adjacencies. Experiments conducted on standard benchmark Bayesian networks indicate that QACD enhances structural coherence and interventional reliability in environments with noisy or inconsistent conditional independence, while still competing effectively with traditional constraint-based, hybrid, and previous argumentation-based methods. This study is available on arXiv in the Computer Science > Artificial Intelligence category.