GENESIS: Towards Explainable Causal Discovery
GENESIS makes each learned causal edge accountable to a stated source of evidence.
The paper defines “decision traceability” as the requirement that every included or excluded edge in a learned DAG be justified by statistical evidence, Markov Blanket consistency, or explicit domain reasoning. GENESIS builds graphs through interpretable steps, starting with scored three-node motifs such as chains, forks, and colliders, then refining with observational evidence. Domain knowledge is invoked only when the statistical signal is insufficient. In experiments, the authors report 100% decision traceability, better SHD than purely statistical methods on most benchmarks, and performance comparable to other LLM-assisted approaches. ArXiv · AI/CL/LG's note
The paper defines “decision traceability” as the requirement that every included or excluded edge in a learned DAG be justified by statistical evidence, Markov Blanket consistency, or explicit domain reasoning. GENESIS builds graphs through interpretable steps, starting with scored three-node motifs such as chains, forks, and colliders, then refining with observational evidence. Domain knowledge is invoked only when the statistical signal is insufficient. In experiments, the authors report 100% decision traceability, better SHD than purely statistical methods on most benchmarks, and performance comparable to other LLM-assisted approaches. ArXiv · AI/CL/LG's note
score 4