[Paper Review] Enabling AI-Generated Content (AIGC) Services in Wireless Edge Networks

Hongyang Du, Zonghang Li+6 more|arXiv (Cornell University)|Jan 9, 2023

Advanced MIMO Systems Optimization45 citations

TL;DR

The paper proposes AIGC-as-a-Service (AaaS) for wireless edge networks, introduces image-based quality metrics and a general model linking resource use to perceived quality, and presents a DRL-based ASP selection strategy to maximize utility while avoiding overloads.

ABSTRACT

Artificial Intelligence-Generated Content (AIGC) refers to the use of AI to automate the information creation process while fulfilling the personalized requirements of users. However, due to the instability of AIGC models, e.g., the stochastic nature of diffusion models, the quality and accuracy of the generated content can vary significantly. In wireless edge networks, the transmission of incorrectly generated content may unnecessarily consume network resources. Thus, a dynamic AIGC service provider (ASP) selection scheme is required to enable users to connect to the most suited ASP, improving the users' satisfaction and quality of generated content. In this article, we first review the AIGC techniques and their applications in wireless networks. We then present the AIGC-as-a-service (AaaS) concept and discuss the challenges in deploying AaaS at the edge networks. Yet, it is essential to have performance metrics to evaluate the accuracy of AIGC services. Thus, we introduce several image-based perceived quality evaluation metrics. Then, we propose a general and effective model to illustrate the relationship between computational resources and user-perceived quality evaluation metrics. To achieve efficient AaaS and maximize the quality of generated content in wireless edge networks, we propose a deep reinforcement learning-enabled algorithm for optimal ASP selection. Simulation results show that the proposed algorithm can provide a higher quality of generated content to users and achieve fewer crashed tasks by comparing with four benchmarks, i.e., overloading-avoidance, random, round-robin policies, and the upper-bound schemes.

Motivation & Objective

• Provide a comprehensive review of AIGC techniques and their edge-network applications.
• Introduce AIGC-as-a-Service (AaaS) and discuss deployment challenges at the edge.
• Propose image-based perceived quality metrics and a general model linking resources to quality.
• Develop a deep reinforcement learning framework for dynamic ASP selection to maximize utility.

Proposed method

• Survey AIGC techniques and their edge applications to motivate AaaS.
• Define AaaS and articulate deployment challenges in wireless edges.
• Propose image-based quality metrics (BRISQUE, TV, DSS, HaarPSI, MDSI, VIF) and a general four-parameter model linking inference steps to quality.
• Develop a deep reinforcement learning (soft actor-critic) algorithm to dynamically select ASPs under resource constraints.
• Compare DRL-based ASP selection with random, round-robin, overloading-avoidance, and upper-bound benchmarks showing improved reliability and QoS.

Experimental results

Research questions

• RQ1 How can AIGC techniques be effectively deployed at wireless edge networks as AaaS?
• RQ2 What are suitable image-based metrics to evaluate AIGC perceived quality without references?
• RQ3 How can resource usage (e.g., inference steps) be modeled to predict perceived content quality in AaaS?
• RQ4 Can DRL-based ASP selection maximize system utility while avoiding overload and task crashes?

Key findings

• A DRL-enabled ASP selection policy achieves zero task crashes and reduces congestion penalties compared with benchmarks.
• The DRL policy rapidly learns load balancing and quality-prioritization, approaching upper-bound performance.
• image-based metrics (BRISQUE, TV, DSS, HaarPSI, MDSI, VIF) correlate with perceived quality under diffusion-model-based AaaS.
• A general four-parameter model can describe the relationship between inference steps and image quality across AaaS types.
• Simulation with 20 ASPs and 1000 edge users demonstrates DRL effectiveness over random, round-robin, and overload-avoidance baselines.