AMATA: Adaptive Multi-Agent Trajectory Alignment for Knowledge-Intensive Question Answering

In three linesAMATA is an adaptive multi-agent trajectory alignment framework for knowledge-intensive question answering. Six specialized agents collaboratively perform structured actions to improve factual consistency and reduce hallucinations. The system formalizes multi-agent collaboration as a trajectory preference alignment problem with intra-trajectory and inter-agent dependency learning.

Summary generated by Claude — human-verified