Incentivizing Multimedia Data Acquisition for Machine Learning System

• Published in: Algorithms and Architectures for Parallel Processing Vol. 11336; pp. 142 - 158
• Main Authors: Gu, Yiren, Shen, Hang, Bai, Guangwei, Wang, Tianjing, Tong, Hai, Hu, Yujia
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2018; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Auction; Incentive mechanism; Machine learning; Multimedia crowdsensing; QoI
• ISBN: 3030050564; 9783030050566
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-05057-3_11

To address restrictions on data collection, incentivizing multimedia data acquisition for machine learning system is proposed. This paper presents an effective QoI (Quality-of-Information)-aware incentive mechanism in multimedia crowdsensing, with the objective of promoting the growth of an initial training model. Firstly, an incentive model is constructed in the form of reverse auction to maximize the social welfare while meeting the requirements in quality, timeliness, correlation and coverage. Then, we discuss how to achieve the optimal social welfare in the presence of an NP-hard winner determination problem. Lastly, a practical incentive mechanism to solve the auction problem is designed, which is shown to be truthful, individually rational and computationally efficient. Extensive simulation results demonstrate the proposed incentive mechanism produces close-to-optimal social welfare noticeably and high-QoI dataset is obtained. In particular, a significant performance improvement for machine learning model growth is achieved with lower complexity.