Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diffusion-based Dynamic Contract for Federated AI Agent Construction in Mobile Metaverses

Jinbo Wen, Jiawen Kang, Yang Zhang, Yue Zhong, Dusit Niyato, Jie Xu, Jianhang Tang, Chau Yuen

TL;DR

This work tackles latency and data-leakage challenges in AI agent construction for mobile metaverses by proposing an edge-cloud federated framework in which edge servers assemble agent modules and a cloud server integrates them for edge deployment. To sustain ES participation under dynamic information asymmetry, the authors design a two-period dynamic contract and solve it with an EDMSAC algorithm that combines diffusion-based policy learning with dynamic structured pruning. The approach integrates energy- and utility-aware models, derives tractable IR/IC conditions, and demonstrates via simulations that EDMSAC achieves superior cloud profit and more reliable contract generation than static or heuristic baselines. The framework enables low-latency, privacy-preserving AI agents in mobile metaverses and provides a scalable path toward adaptive, multi-period incentive mechanisms for heterogeneous edge resources.

Abstract

Mobile metaverses have attracted significant attention from both academia and industry, which are envisioned as the next-generation Internet, providing users with immersive and ubiquitous metaverse services through mobile devices. Driven by Large Language Models (LLMs) and Vision-Language Models (VLMs), Artificial Intelligence (AI) agents hold the potential to empower the creation, maintenance, and evolution of mobile metaverses. Currently, AI agents are primarily constructed using cloud-based LLMs and VLMs. However, several challenges hinder their effective implementation, including high service latency and potential sensitive data leakage during perception and processing. In this paper, we develop an edge-cloud collaboration-based federated AI agent construction framework in mobile metaverses. Specifically, Edge Servers (ESs), acting as agent infrastructures, collaboratively create agent modules in a distributed manner. The cloud server then integrates these modules into AI agents and deploys them at the edge, thereby enabling low-latency AI agent services for users. Considering that ESs may exhibit dynamic levels of willingness to participate in federated AI agent construction, we design a two-period dynamic contract model to continuously motivate ESs to participate in agent module creation, effectively addressing the dynamic information asymmetry between the cloud server and the ESs. Furthermore, we propose an Enhanced Diffusion Model-based Soft Actor-Critic (EDMSAC) algorithm to efficiently generate optimal dynamic contracts, in which dynamic structured pruning is applied to DM-based actor networks to enhance denoising efficiency and policy learning performance. Extensive simulations demonstrate the effectiveness and superiority of the EDMSAC algorithm and the proposed contract model.

Diffusion-based Dynamic Contract for Federated AI Agent Construction in Mobile Metaverses

TL;DR

This work tackles latency and data-leakage challenges in AI agent construction for mobile metaverses by proposing an edge-cloud federated framework in which edge servers assemble agent modules and a cloud server integrates them for edge deployment. To sustain ES participation under dynamic information asymmetry, the authors design a two-period dynamic contract and solve it with an EDMSAC algorithm that combines diffusion-based policy learning with dynamic structured pruning. The approach integrates energy- and utility-aware models, derives tractable IR/IC conditions, and demonstrates via simulations that EDMSAC achieves superior cloud profit and more reliable contract generation than static or heuristic baselines. The framework enables low-latency, privacy-preserving AI agents in mobile metaverses and provides a scalable path toward adaptive, multi-period incentive mechanisms for heterogeneous edge resources.

Abstract

Mobile metaverses have attracted significant attention from both academia and industry, which are envisioned as the next-generation Internet, providing users with immersive and ubiquitous metaverse services through mobile devices. Driven by Large Language Models (LLMs) and Vision-Language Models (VLMs), Artificial Intelligence (AI) agents hold the potential to empower the creation, maintenance, and evolution of mobile metaverses. Currently, AI agents are primarily constructed using cloud-based LLMs and VLMs. However, several challenges hinder their effective implementation, including high service latency and potential sensitive data leakage during perception and processing. In this paper, we develop an edge-cloud collaboration-based federated AI agent construction framework in mobile metaverses. Specifically, Edge Servers (ESs), acting as agent infrastructures, collaboratively create agent modules in a distributed manner. The cloud server then integrates these modules into AI agents and deploys them at the edge, thereby enabling low-latency AI agent services for users. Considering that ESs may exhibit dynamic levels of willingness to participate in federated AI agent construction, we design a two-period dynamic contract model to continuously motivate ESs to participate in agent module creation, effectively addressing the dynamic information asymmetry between the cloud server and the ESs. Furthermore, we propose an Enhanced Diffusion Model-based Soft Actor-Critic (EDMSAC) algorithm to efficiently generate optimal dynamic contracts, in which dynamic structured pruning is applied to DM-based actor networks to enhance denoising efficiency and policy learning performance. Extensive simulations demonstrate the effectiveness and superiority of the EDMSAC algorithm and the proposed contract model.

Paper Structure

This paper contains 31 sections, 14 theorems, 77 equations, 9 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Federated Model Construction
  4. Contract Theory-based Applications
  5. Deep Reinforcement Learning for Contract Design
  6. Edge-Cloud Collaboration-based Federated AI Agent Construction Framework
  7. Framework Design
  8. Model Quality of Agent Modules
  9. Energy Consumption of Edge Servers
  10. Utilities of Edge Servers and Cloud Server
  11. Dynamic Contract Design
  12. Two-Period Dynamic Contract Formulation
  13. Second-Period Contract Design
  14. First-Period Contract Design
  15. Optimal Two-Period Contract
  16. ...and 16 more sections

Key Result

Lemma 1

For any feasible contract, $R_i^2(\theta_k^1) \geq R_j^2(\theta_k^1)$ if and only if $T_i^2(\theta_k^1) \geq T_j^2(\theta_k^1)$, where $i\neq j$.

Figures (9)

  • Figure 1: The proposed edge-cloud collaboration-based federated AI agent construction framework. Note that the interface of personal assistants is from the OmAgent app zhang2024omagentmultimodalagentframework downloaded by our Android phone.
  • Figure 2: The workflow of the proposed two-period dynamic contract for agent module construction. To minimize disruptions to the construction progress, the proposed contract remains valid across both periods without requiring redefinition and retransmission to the ESs in period 2.
  • Figure 3: The architecture of the proposed EDMSAC algorithm for dynamic contract design. The EDMSAC algorithm can generate optimal two-period contracts through the denoising process. To enhance denoising efficiency, we innovatively apply dynamic structured pruning techniques to the MLP of DM-based actor networks. Note that the unimportant neurons in the actor networks are simply masked, not removed outright.
  • Figure 4: Two case studies of mobile phone operations using an AI agent. We call the Qwen-VL-Plus and Qwen-VL-Max models through the Qwen-VL API to construct the mobile AI agent, utilizing Android Debug Bridge (ADB) to operate a mobile phone with the Android OS system, where ADB can simulate all operations of the AI agent wang2024mobileagentv2mobiledeviceoperation.
  • Figure 5: Performance evaluation of mobile AI agents.
  • ...and 4 more figures

Theorems & Definitions (26)

  • Definition 1
  • Definition 2
  • Lemma 1
  • proof
  • Lemma 2
  • proof
  • Lemma 3
  • Lemma 4
  • proof
  • Lemma 5
  • ...and 16 more