Toward Incentive With Privacy Preserving Machine Learning as a Service for Crowdsensed Data Trading

With the popularization and development of artificial intelligence technology, as well as the increasingly deep integration with various industries, machine learning as a service model is gradually gaining popularity and maturing. However, in the process of model sharing services, there is still data privacy leakage, which poses security risks to data usage security. To address this challenge, this paper proposes a towards incentive with privacy preserving machine learning as a service scheme for crowdsensed data trading. This scheme converts the data sharing problem into a federated learning model sharing problem, and then converts the shared model into an auction model, thereby achieving the transformation of privacy protection issues during the sharing process into privacy auction problems. In auction mode, while ensuring the security of submitted information, characteristics such as utility, individually rational and maximizing social welfare need to be met. Furthermore, in order to ensure fairness and privacy, the bidding information sorting algorithm and pricing strategy under the ciphertext state are designed. Once the winners are determined, the model service sharing mode based on attribute-based encryption and InterPlanetary File System is adopted. The extended experimental results indicate that the proposed scheme meets the characteristics of privacy preserving, flexibility, and efficiency.