Background: Some of the new challenges that consumers are confronted with include safeguarding data and privacy. Data retrieved from the cloud's external sources and the calculations that follow aren't always accurate. Addressing the risk of adversaries utilizing various exploitation techniques to compromise transactional data privacy is a basic concern when establishing big private networks. The profession of criminal investigation is quickly embracing new technologies, and one of these is blockchain. Every sector, from banking and supply chain management to smart apps and the Internet of Things (IoT), has been increasingly vulnerable to security threats in recent years. Methods: An effective solution to the "data island" problem, federated-learning (FL) has recently been a hot and broad concern topic. But as FL technology finds more practical uses, training management gets more complicated, and the trade-off of multi-tasking gets more difficult, due to the increasing quantity of FL tasks. A privacy-preserving FL framework with multi-tasks using a partitioned blockchain is proposed in this study to address this shortcoming. The framework may execute several FL tasks by separate requesters. To start, an FL task force is established to help with the visualization, organization, and administration of security aggregation. Result: In order to safeguard users' privacy and guarantee the accuracy of the global model, the suggested framework incorporates both Paillier-homomorphic-encryption(PHE) and Pearson-correlation-coefficient(PCC). Lastly, a novel incentive system based on the blockchain is introduced to encourage individuals to provide their valuable data. Conclusion: Our suggested framework achieves a global model accuracy of 99.2% according to the experimental data. Specifically, in the realm of industrial applications, the suggested framework is clearly more suited to real-world settings.
Read full abstract