研究库 论文知识库

内容库 / 论文

arXiv:2609.18779 · Agent 智能体

CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM Agents

CERA-MoA：与持续学习 LLM Agent 协同进化的路由机制

CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM Agents

类型：arxiv
标识：2609.18779
链接：https://arxiv.org/abs/2609.18779
主分类：agent
形态：method
TLDR：Current Mixture-of-Agents (MoA) paradigms generally treat query routing and agent fine-tuning as separate processes, limiting their ability to respond to evolving agent capabilities. This disconnect prevents routing strategies from adapting to evolving agent capabilities during post-training and prevents agents from achieving synergistic data-driven specialization. To resolve this, we introduce CERA-MoA (Co-Evolving Router with continually learning Agents for Mixture-of-Agents), an iterative reinforcement learning framework where the dynamic router and independent agent policies co-evolve. We
副分类：engineering
待LLM分类：否
标题中文：CERA-MoA：与持续学习 LLM Agent 协同进化的路由机制
TLDR中文：当前 Mixture-of-Agents（MoA）范式通常将查询路由与 Agent 微调视为分离过程，限制了其响应 Agent 能力演进的灵活性。这种割裂使路由策略无法在后训练期间适应 Agent 能力变化，也使 Agent 无法实现数据驱动的协同特化。为解决此问题，我们提出 CERA-MoA（Mixture-of-Agents 中与持续学习 Agent 协同进化的路由器），一个迭代强化学习框架，其中动态路由器与独立 Agent 策略协同进化。我们...
来源文件：
/inbox/tom/_candidates/2026-09-17-agent-rag-longcontext-candidates.json

相关 · 按语义匹配

笔记深度精读 · SoK: Agentic Retrieval-Augmented Generation（arXiv:2603.07379）

笔记 Tom 文献雷达 · Agent + RAG + 长上下文 · 2026-08-27 20:40

笔记 Tom 文献雷达 · Agent/RAG/长上下文 · 2026-08-03 08:40 UTC

笔记 Tom 文献雷达 · Agent × RAG × 长上下文 · 2026-07-27 20:40 UTC

笔记 📋 Tom 文献雷达 · Agent + RAG + 长上下文 · 2026-09-01

笔记 Tom 文献雷达 · Agent + RAG + LongContext · 2026-08-29

arXiv

2609.18779

TLDR

当前 Mixture-of-Agents（MoA）范式通常将查询路由与 Agent 微调视为分离过程，限制了其响应 Agent 能力演进的灵活性。这种割裂使路由策略无法在后训练期间适应 Agent 能力变化，也使 Agent 无法实现数据驱动的协同特化。为解决此问题，我们提出 CERA-MoA（Mixture-of-Agents 中与持续学习 Agent 协同进化的路由器），一个迭代强化学习框架，其中动态路由器与独立 Agent 策略协同进化。我们...Current Mixture-of-Agents (MoA) paradigms generally treat query routing and agent fine-tuning as separate processes, limiting their ability to respond to evolving agent capabilities. This disconnect prevents routing strategies from adapting to evolving agent capabilities during post-training and prevents agents from achieving synergistic data-driven specialization. To resolve this, we introduce CERA-MoA (Co-Evolving Router with continually learning Agents for Mixture-of-Agents), an iterative reinforcement learning framework where the dynamic router and independent agent policies co-evolve. We

arXiv ↗ 中文全文