Tag
This paper proposes a framework that ensembles the reasoning structures of multiple LLMs by weighted merging of extracted Directed Acyclic Graphs (DAGs), enabling consensus reasoning with improved accuracy and interpretability across several benchmarks.