Abstract:A novel accelerated mobile edge generation (MEG) framework is proposed for generating high-resolution images on mobile devices. Exploiting a large-scale latent diffusion model (LDM) distributed across edge server (ES) and user equipment (UE), cost-efficient artificial intelligence generated content (AIGC) is achieved by transmitting low-dimensional features between ES and UE. To reduce overheads of both distributed computations and transmissions, a dynamic diffusion and feature merging scheme is conceived. By jointly optimizing the denoising steps and feature merging ratio, the image generation quality is maximized subject to latency and energy consumption constraints. To address this problem and tailor LDM sub-models, a low-complexity MEG acceleration protocol is developed. Particularly, a backbone meta-architecture is trained via offline distillation. Then, dynamic diffusion and feature merging are determined in online channel environment, which can be viewed as a constrained Markov Decision Process (MDP). A constrained variational policy optimization (CVPO) based MEG algorithm is further proposed for constraint-guaranteed learning, namely MEG-CVPO. Numerical results verify that: 1) The proposed framework can generate 1024$\times$1024 high-quality images over noisy channels while reducing over $40\%$ latency compared to conventional generation schemes. 2) The developed MEG-CVPO effectively mitigates constraint violations, thus flexibly controlling the trade-off between image distortion and generation costs.

What problem does this paper attempt to address?

The main problem this paper attempts to address is how to accelerate the generation of high-resolution images within the Mobile Edge Generation (MEG) framework while reducing computational and transmission overhead, and meeting latency and energy consumption constraints. Specifically, the paper proposes a novel accelerated MEG framework by distributing large-scale Latent Diffusion Models (LDM) between edge servers (ES) and user equipment (UE), transmitting only low-dimensional features to achieve efficient Artificial Intelligence Generated Content (AIGC). To further reduce computational and transmission overhead, the paper designs a dynamic diffusion and feature compression scheme, and maximizes image quality through joint optimization of denoising steps and feature fusion ratios. Additionally, the paper proposes a MEG algorithm based on Constrained Variational Policy Optimization (CVPO) (MEG-CVPO) to solve the resulting constrained Markov Decision Process (MDP), ensuring the satisfaction of constraints and achieving a flexible trade-off between image quality and generation cost. ### Main Contributions: 1. **Proposed a novel accelerated MEG framework**: By distributing large-scale LDM between UE and ES, achieving efficient and low-cost high-resolution image generation. 2. **Developed a low-complexity dynamic MEG acceleration protocol**: First, the backbone architecture of the LDM sub-model is learned through offline distillation, then in an online environment, the denoising steps and feature fusion ratios are dynamically predicted based on channel observations, reformulating the problem as a constrained MDP. 3. **Proposed a CVPO-based MEG algorithm (MEG-CVPO)**: By training strategies through variational inference, it improves system rewards while achieving feasible policy distribution, effectively mitigating constraint violation issues, and solving the constrained MDP. 4. **Provided numerical results to verify effectiveness**: Compared to traditional generation schemes, the proposed dynamic diffusion and feature fusion scheme can reconstruct high-quality 1024×1024 images through noisy channels within 3-7 seconds, reducing latency by over 40%, and the MEG-CVPO algorithm outperforms traditional Lagrangian methods in constraint assurance, achieving a controllable trade-off between generation quality and cost.

Accelerating Mobile Edge Generation (MEG) by Constrained Learning

Enabling Distributed Generative Artificial Intelligence in 6G: Mobile Edge Generation

Mobile Edge Generation: A New Era to 6G

Mobile Edge Generation-Enabled Digital Twin: Architecture Design and Research Opportunities

MIEGAN: Mobile Image Enhancement Via a Multi-Module Cascade Neural Network.

Learning for Exception: Dynamic Service Caching in 5G-Enabled MECs with Bursty User Demands

Energy-Efficient Incremental Offloading of Neural Network Computations in Mobile Edge Computing

Toward Scalable Generative AI via Mixture of Experts in Mobile Edge Networks

Adaptive Progressive Image Enhancement for Edge-Assisted Mobile Vision

MobileMEF: fast and efficient method for real-time mobile multi-exposure fusion

Accelerating Convergence of Federated Learning in MEC with Dynamic Community

Latency-Aware Resource Allocation for Mobile Edge Generation and Computing via Deep Reinforcement Learning

Reinforcement Learning Based Energy-Efficient Collaborative Inference for Mobile Edge Computing

Retrieval-Augmented Generation for Mobile Edge Computing via Large Language Model

Energy-Efficient Collaborative Inference in MEC: A Multi-Agent Reinforcement Learning Based Approach

Efficient Transformer Inference for Extremely Weak Edge Devices Using Masked Autoencoders.

Accelerated Image-Aware Generative Diffusion Modeling

Optimal Design of Hybrid Federated and Centralized Learning in the Mobile Edge Computing Systems

Cross-Modal Generative Semantic Communications for Mobile AIGC: Joint Semantic Encoding and Prompt Engineering

Delay-Aware Parallel Offloading AIGC Service in Edge Computing

MEC-Enabled Wireless VR Video Service: A Learning-Based Mixed Strategy for Energy-Latency Tradeoff.