Big Data Analytics by CrowdLearning: Architecture and Mechanism Design
Yufeng Zhan, Peng Li, Kun Wang, Song Guo, Yuanqing Xia
Abstract
Crowdsensing has emerged as a powerful tool to collect IoT big data. Moving big data to the cloud for analysis is time consuming and has the risk of data privacy leakage. An alternative is to leave the training data distributed on mobile devices, and learn a shared model by aggregating locally computed updates. In this article, we propose a CrowdLearning system, which employs MUs for big data collection and deep learning training. We propose a game-based incentive mechanism to optimize the utilities of MUs and accuracy of the training model by exploiting the various sensing and training capabilities of MUs. Experiments have been conducted to evaluate the performance of proposed CrowdLearning system and the results validate the effectiveness of the proposed mechanism.