1.《Blockchain and Federated Learning for Privacy-Preserved Data Sharing in Industrial IoT》
In this article, the author proposes a different private multi-party data model sharing method based on permissioned blockchain . We did not directly share the original data, but combined the federated learning algorithm to map the original data to the corresponding data model. Through distributed training by local users, the privacy problem in the learning phase was solved. We also designed a distributed data sharing architecture based on the blockchain , so that the blockchain can guarantee data retrieval and ensure accurate training of the model. The main contributions in this article are as follows:
1) By using federated learning to establish a data model, sharing the data model instead of the original data, the data sharing problem is transformed into a machine learning problem.
2) A new blockchain authorization collaboration architecture is proposed, which reduces the risk of data leakage through distributed multi-party sharing of data, and data owners can further control access to shared data.
3) Integrate differentiated privacy into federal learning to further protect data privacy. We also evaluated the effectiveness and benchmarks of our proposed model, the data classification of open real data sets.
2.《Blockchain-based Node-aware Dynamic Weighting Methods for Improving Federated Learning Performance》
A federal learning scenario based on blockchain is proposed.
The local learning weighting method based on node recognition, the method of selecting nodes according to the participation frequency and data volume, and the method of weighting according to the participation frequency are proposed to quickly stabilize the convergence learning accuracy. It also compares and analyzes the difference between this method and the traditional federated learning method in terms of learning speed and stability. The results show that the method is superior to the traditional federated learning method in terms of learning speed and stability.
3.《FLchain: Federated Learning via MEC-enabled Blockchain Network》
A blockchain network-based architecture called "FLchain" is proposed to enhance the security of Federated Learning (FL).
Use the concept of channels to learn multiple global models on FLchain. The local model parameters for each global iteration are stored as blocks in the channel-specific ledger.
The author combines multi-access edge computing (MEC) with blockchain network and proposes a system model suitable for FL. The physical structure of FLchain includes mobile devices and edge devices.
The mobile device uses the data samples on the device to calculate local model updates.
Edge devices serve two purposes. 1) They provide network resources to mobile devices with limited resources. 2) They act as nodes in FLchain's blockchain network.