Federal learning method based on fog calculation

Abstract

The invention relates to a federal learning method based on fog calculation, and belongs to the technical field of mobile communication. An operator located at the cloud serves as a publisher of a federated learning task, a fog node located at the edge serves as a block chain consensus node to provide safe coordination service for federated learning, and meanwhile, Internet of Things equipment such as a smart phone and a tablet personal computer serve as a client of federated learning. Through the security verification service provided by the block chain and a given clustering federated learning method, the federated learning can effectively improve the federated learning efficiency and stability. In order to solve a single-point fault problem and a model hostile attack problem existing in federated learning, a client collaborative learning mode based on a block chain is provided, in order to reduce extra time delay generated by introduction of a block chain technology, a block chain consensus mode maintained by a fog node is PBFT, a block chain network is divided into a plurality of sub-networks, and consensus verification is carried out based on different federated learning tasks in each sub-network to reduce consensus time delay.

Description

technical field [0001] The invention belongs to the technical field of mobile communication, and relates to a federated learning method based on fog computing. Background technique [0002] At present, the issue of privacy protection in machine learning is particularly important, and private data generated by users should not be exposed or uploaded to central servers. Google proposed Vanilla Federation Learning (FL) in 2016 to solve the problem of collaborative training for privacy protection. FL is a decentralized framework to collaboratively learn models using training data distributed across remote devices to improve communication efficiency. Basically, it learns a shared pre-trained model by aggregating model updates from FL participating devices (locally computed model updates based on training data distributed across participating devices). The aggregation algorithm used in VanillaFL is responsible for averaging the parameters of many local models. [0003] Due to t...

AI Technical Summary

Problems solved by technology

However, in FL, due to data heterogeneity caused by data sources and computing nodes being personal devices with different owners and network conditions, also known as Non-IID data, a general global model often cannot adapt to all users The generated data distribution, which is inconsistent with the implicit assumptions of FL, leads to insufficient expressive power of the global model

Moreover, with the increase of malicious attacks, the security of FL is also decreasing

Images