Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Decentralized Multi-Agent Coordination with Privacy-Preserving Knowledge Sharing: Extending AgentNet for Scalable Autonomous Systems

Goutham Nalagatla

TL;DR

The paper tackles scalability and privacy challenges in fully decentralized LLM-based multi-agent coordination by introducing AgentNet++, a hierarchical framework that organizes agents into clusters, enables privacy-preserving knowledge sharing via differential privacy and secure aggregation, and employs adaptive resource management with theoretical convergence and privacy guarantees. The methodology combines multi-level architecture, cluster-head coordination, and a task-routing score that balances expertise, resources, and load. Theoretical results establish convergence, privacy, and scalability bounds, while experiments show substantial gains in task completion, communication efficiency, and privacy preservation, scalable to 1000+ agents. This work advances practical, private, and scalable emergent intelligence in large-scale autonomous systems and provides open-source implementations for reproducibility.

Abstract

Decentralized multi-agent systems have shown promise in enabling autonomous collaboration among LLM-based agents. While AgentNet demonstrated the feasibility of fully decentralized coordination through dynamic DAG topologies, several limitations remain: scalability challenges with large agent populations, communication overhead, lack of privacy guarantees, and suboptimal resource allocation. We propose AgentNet++, a hierarchical decentralized framework that extends AgentNet with multilevel agent organization, privacy-preserving knowledge sharing via differential privacy and secure aggregation, adaptive resource management, and theoretical convergence guarantees. Our approach introduces cluster-based hierarchies where agents self-organize into specialized groups, enabling efficient task routing and knowledge distillation while maintaining full decentralization. We provide formal analysis of convergence properties and privacy bounds, and demonstrate through extensive experiments on complex multi-agent tasks that AgentNet++ achieves 23% higher task completion rates, 40% reduction in communication overhead, and maintains strong privacy guarantees compared to AgentNet and other baselines. Our framework scales effectively to 1000+ agents while preserving the emergent intelligence properties of the original AgentNet.

Hierarchical Decentralized Multi-Agent Coordination with Privacy-Preserving Knowledge Sharing: Extending AgentNet for Scalable Autonomous Systems

TL;DR

The paper tackles scalability and privacy challenges in fully decentralized LLM-based multi-agent coordination by introducing AgentNet++, a hierarchical framework that organizes agents into clusters, enables privacy-preserving knowledge sharing via differential privacy and secure aggregation, and employs adaptive resource management with theoretical convergence and privacy guarantees. The methodology combines multi-level architecture, cluster-head coordination, and a task-routing score that balances expertise, resources, and load. Theoretical results establish convergence, privacy, and scalability bounds, while experiments show substantial gains in task completion, communication efficiency, and privacy preservation, scalable to 1000+ agents. This work advances practical, private, and scalable emergent intelligence in large-scale autonomous systems and provides open-source implementations for reproducibility.

Abstract

Decentralized multi-agent systems have shown promise in enabling autonomous collaboration among LLM-based agents. While AgentNet demonstrated the feasibility of fully decentralized coordination through dynamic DAG topologies, several limitations remain: scalability challenges with large agent populations, communication overhead, lack of privacy guarantees, and suboptimal resource allocation. We propose AgentNet++, a hierarchical decentralized framework that extends AgentNet with multilevel agent organization, privacy-preserving knowledge sharing via differential privacy and secure aggregation, adaptive resource management, and theoretical convergence guarantees. Our approach introduces cluster-based hierarchies where agents self-organize into specialized groups, enabling efficient task routing and knowledge distillation while maintaining full decentralization. We provide formal analysis of convergence properties and privacy bounds, and demonstrate through extensive experiments on complex multi-agent tasks that AgentNet++ achieves 23% higher task completion rates, 40% reduction in communication overhead, and maintains strong privacy guarantees compared to AgentNet and other baselines. Our framework scales effectively to 1000+ agents while preserving the emergent intelligence properties of the original AgentNet.

Paper Structure

This paper contains 20 sections, 5 equations, 3 figures, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Decentralized Multi-Agent Systems
  4. Hierarchical Multi-Agent Systems
  5. Privacy in Multi-Agent Learning
  6. Methodology
  7. System Architecture
  8. Hierarchical Task Decomposition and Routing
  9. Privacy-Preserving Knowledge Sharing
  10. Adaptive Resource Management
  11. Cluster Formation and Evolution
  12. Theoretical Analysis
  13. Convergence Guarantees
  14. Privacy Guarantees
  15. Scalability Analysis
  16. ...and 5 more sections

Figures (3)

  • Figure 1: Scalability comparison: Execution time vs number of agents. AgentNet++ maintains logarithmic growth while AgentNet and Centralized show polynomial degradation.
  • Figure 2: Task completion rate comparison across different methods. AgentNet++ achieves the highest success rate (87.3%) with lower variance. Error bars represent standard error over 10 runs.
  • Figure 3: Left: Privacy-utility trade-off showing task success rate as a function of privacy budget $\epsilon$. Right: Communication overhead comparison demonstrating AgentNet++'s superior efficiency ($O(n^{1.5})$) vs AgentNet ($O(n^2)$).