Learning-based Big Data Sharing Incentive in Mobile AIGC Networks

Jinbo Wen,Yang Zhang,Yulin Chen,Weifeng Zhong,Xumin Huang,Lei Liu,Dusit Niyato

2024-08-01

Abstract:Rapid advancements in wireless communication have led to a dramatic upsurge in data volumes within mobile edge networks. These substantial data volumes offer opportunities for training Artificial Intelligence-Generated Content (AIGC) models to possess strong prediction and decision-making capabilities. AIGC represents an innovative approach that utilizes sophisticated generative AI algorithms to automatically generate diverse content based on user inputs. Leveraging mobile edge networks, mobile AIGC networks enable customized and real-time AIGC services for users by deploying AIGC models on edge devices. Nonetheless, several challenges hinder the provision of high-quality AIGC services, including issues related to the quality of sensing data for AIGC model training and the establishment of incentives for big data sharing from mobile devices to edge devices amidst information asymmetry. In this paper, we initially define a Quality of Data (QoD) metric based on the age of information to quantify the quality of sensing data. Subsequently, we propose a contract theoretic model aimed at motivating mobile devices for big data sharing. Furthermore, we employ a Proximal Policy Optimization (PPO) algorithm to determine the optimal contract. Numerical results demonstrate the efficacy and reliability of the proposed PPO-based contract model.

Networking and Internet Architecture

What problem does this paper attempt to address?

The paper primarily investigates the incentive mechanism for big data sharing in mobile Artificial Intelligence Generated Content (AIGC) networks. Specifically, the paper addresses the following key issues:

1. **Data Quality Quantification**: To enhance the effectiveness of AIGC model training, the paper proposes a quality metric based on the Age of Information (AoI), known as the Quality of Data (QoD), to quantify the quality of sensor data. This metric considers service latency and the freshness of service data to ensure the timeliness and effectiveness of the training data.

2. **Incentive Mechanism Design**: Due to the information asymmetry when mobile devices contribute data, they may be unwilling or maliciously provide low-quality data. The paper establishes an incentive model through contract theory to encourage mobile devices to provide high-quality data. This model aims to maximize the overall utility of the edge server while considering the impact of information asymmetry.

3. **Decision Process Optimization**: Given the high-dimensional nature of the mobile AIGC network environment, traditional mathematical methods find it challenging to identify the optimal strategy. Therefore, the paper employs the Proximal Policy Optimization (PPO) algorithm to determine the optimal contract design. This method effectively handles dynamically changing states and demonstrates good performance in experiments.

In summary, by defining the QoD metric, designing an incentive mechanism, and adopting the PPO algorithm, the paper aims to address the three core challenges of how to quantify sensor data quality, how to incentivize mobile devices to provide high-quality data, and how to handle complex decision processes. The ultimate goal is to improve the quality and reliability of AIGC services.