MANTA: Multi-Agent Network Topology Adaptation for Self-Evolving Multi-Agent Systems

Source

Source name: arXiv
Source kind: paper
Authors / organization: Mao-xun Huang, Jerry Wang, Yi-Cheng Lai, Zhengxin Zhang, Claire Cardie, Hen-Hsen Huang
Venue: arXiv: cs.AI
Published: 2026-07-30
Collected: 2026-07-31
URL: http://arxiv.org/abs/2607.28527v1
PDF: https://arxiv.org/pdf/2607.28527v1
DOI: UNKNOWN

Plain-Language Brief

The paper introduces MANTA, a framework for Multi-Agent Network Topology Adaptation that allows communication structures in large language model-based multi-agent systems to evolve during deployment. The goal is to improve complex problem solving by adapting agent roles, communication links, execution order, information visibility, and validation pathways while preserving the task interface and agent budget.

What the Document Says

The paper presents MANTA, a framework for Multi-Agent Network Topology Adaptation.
MANTA enables self-evolving communication structures at inference time.
Before execution, MANTA initializes a task-conditioned topology from prior structural experience.
During deployment, it monitors collaboration traces and applies bounded structural updates when the current organization becomes insufficient.
These updates can modify agent roles, communication links, execution order, information visibility, and validation pathways.
The goal is to preserve the task interface and agent budget.
The paper evaluates MANTA against representative single-agent and multi-agent baselines on five benchmarks.
MANTA achieves the highest average score of 74.0, outperforming the strongest baseline by 5.8 percentage points.

Key Concepts Mentioned

Multi-Agent Systems
Large Language Models
Task Decomposition
Agent Specialization
Information Exchange
Inference Time
Communication Topology
Self-Evolving
Collaboration Traces
Bounded Structural Updates
Agent Roles
Communication Links
Execution Order
Information Visibility
Validation Pathways
Task Interface
Agent Budget
Benchmarks

Claims Made by the Document

MANTA improves complex problem solving by adapting the architecture of collaboration.
MANTA outperforms the strongest baseline by 5.8 percentage points.

Evidence Used

Results from evaluations against representative baselines on five benchmarks.

Methods / Approach

Task-conditioned topology initialization, Collaboration trace monitoring, Bounded structural updates

Limitations Stated by the Authors

Not stated in the source.

Limitations Noted by the Agent

Agent interpretation — not from the source.

The performance of MANTA might be influenced by factors not considered in this study.

Open Questions

What are the specific factors that could influence the performance of MANTA?
How does MANTA handle complex scenarios where multiple agents need to collaborate simultaneously?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The paper does not discuss potential limitations or challenges in detail, so further questions have been generated based on common concerns in similar research.

Raw Abstract / Excerpt

Large language model-based multi-agent systems improve complex problem solving through task decomposition, agent specialization, information exchange, and intermediate validation. However, existing systems typically treat communication topology as a fixed design choice or an offline optimization target. We introduce MANTA, a framework for Multi-Agent Network Topology Adaptation that enables communication structures to self-evolve at inference time. Before execution, MANTA initializes a task-conditioned topology from prior structural experience. During deployment, it monitors collaboration traces and applies bounded structural updates when the current organization becomes insufficient. These updates can modify agent roles, communication links, execution order, information visibility, and validation pathways while preserving the task interface and agent budget. We evaluate MANTA against representative single-agent and multi-agent baselines on five benchmarks spanning information seeking, tool use, planning, workflow execution, and mathematical reasoning. MANTA achieves the highest average score of 74.0, outperforming the strongest baseline by 5.8 percentage points and obtaining the best result on PlanCraft. These results show that inference-time self-improvement can extend to the architecture of collaboration itself.