Federated learning defense method based on block chain

Abstract

The invention discloses a federated learning defense method based on a block chain. The method comprises the following steps: a participant establishing an intelligent contract with an authority; obtaining the model from the block chain by the participant in the book, carrying out local training, uploading the trained local model and the corresponding training time to the corresponding block node,and broadcasting the local model and the corresponding training time to the block chain; constructing a noise committee of a noise committee for each on-book participant, and adding noise to the local model of the corresponding on-book participant by using the noise committee to update the local model to obtain an updated model; constructing a verification committee for all in-book participants,verifying the prediction reliability and authenticity of each update model by using the verification committee according to the data set and the training time, and recording the update model passing the verification in a new block node; and the authority obtaining all the updated models passing the verification from the block nodes and aggregating the updated models to obtain an aggregated model,and broadcasting the aggregated model to the block chain for the next round of in-book participants to download local training.

Description

technical field [0001] The invention belongs to the fields of machine learning, federated learning and blockchain, and in particular relates to a blockchain-based federated learning defense method. Background technique [0002] Google proposed Federated Learning (FL) in 2016, using distributed technology to solve extensive research problems in the deep learning environment in recent years. FL has a distributed training model with two roles: participating devices and a central server. Instead of uploading private data, nodes update the global model locally and then upload model updates (or local gradient updates). A central server collects these updates and integrates them to form an updated model. This training process repeats until the training error is less than a pre-specified threshold. In recent years, due to the privacy of FL, a subtle threat has been posed: clients who previously acted as passive data providers are now actively participating in the training process...

AI Technical Summary

Problems solved by technology

[0003] For the privacy of federated learning, there are currently two main challenges: (1) Malicious attackers can attack the shared model through poisoning attacks, where the attacker provides corresponding updates to affect the shared model parameters, while maintaining the global accuracy, make the model misclassify on some tasks

(2) Malicious attackers can also attack the privacy of other customers

However, a private decentralized solution that addresses both attacks simultaneously will not work

Images