[论文] SDOF: Taming the Alignment Tax in Multi-Agent Orchestration with State...

论文概要

研究领域: ML 作者: Zhantao Wang 发布时间: 2025-05-15 arXiv: 2505.10891

中文摘要

LangChain、LangGraph和CrewAI等多智能体编排框架通过基于图的管道路由任务，但并不强制执行真实业务流程中的阶段约束。我们提出SDOF，一种将多智能体执行视为约束状态机的框架。SDOF通过三个组件实现两个主要防御层：（1）通过生成式奖励建模（GRPO）训练的在线RLHF专用意图路由器，以及（2）带GoalStage有限自动机检查和前置/后置条件SkillRegistry验证的StateAwareDispatcher，实现可审计的执行控制。在由北森iTalent平台（6000+企业）支持的招聘系统上，185个专家策划的场景触发1671次实时API调用。我们的GSPO对齐7B意图路由器在该FSM约束对抗路由基准上，联合准确率超过零样本GPT-4o（80.9% 对 48.9%）。端到端执行中，SDOF达到86.5%任务完成率（95%置信区间80.8-90.7），并阻止了注入、非法HR子集中的全部22次操作。在更广泛的消息级阻断审计中，SDOF达到精确率100%、召回率88%，专家一致性kappa=0.94。在覆盖8个服务领域的960个SGD派生对话的单独评估中，我们的FSM映射下浮现201个阶段顺序冲突，其中41个出现在正常分组中。

原文摘要

Multi-agent orchestration frameworks such as LangChain, LangGraph, and CrewAI route tasks through graph-based pipelines but do not enforce the stage constraints that govern real business processes. We present SDOF, a framework that treats multi-agent execution as a constrained state machine. SDOF operates through two primary defensive layers, implemented by three components: (1) an Online-RLHF Specialized Intent Router trained via Generative Reward Modeling (GRPO) and (2) a StateAwareDispatcher with GoalStage finite-automaton checks and precondition/postcondition SkillRegistry validation for auditable execution control. On a recruitment system backed by the Beisen iTalent platform (6000+ enterprises), 185 expert-curated scenarios trigger 1671 live API calls. Our GSPO-aligned 7B Intent Rout...

自动采集于 2026-05-19

#论文 #arXiv #ML #小凯