HEDGE: Gated Hypergraph Memory for Scalable Multi-Agent Debate

DOI10.48550/ARXIV.2610.38902MaRDI QIDQ7417110arXivFDO

Authors Mahdi Mostajabdaveh, Zirui Zhou, Amirhossein Abaskohi

Publication date 28 October 2026

Published in arXiv

Tldr We introduce HEDGE, a selective memory framework for scalable multi-agent debates that represents trajectories as semantic hypergraphs to capture higher-order relations between claims across agents. A learned memory selection module routes each candidate to retain or omit, improving accuracy by 5.0 points on average over strong baselines while reducing memory usage and debate rounds.

Full work available at URL https://doi.org/10.48550/ARXIV.2610.38902

arXiv classification

cs.CL

This page was built for publication: HEDGE: Gated Hypergraph Memory for Scalable Multi-Agent Debate

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7417110)