Federated learning toilet vehicle attack defense method based on block chain

Abstract

The invention discloses a federated learning toilet vehicle attack defense method based on a block chain, and the method comprises the steps: building an intelligent contract between a task publishing mechanism and a client in the block chain according to a calculation capability proof, and achieving the first screening and filtering of a toilet vehicle attack client based on the intelligent contract; enabling the task issuing mechanism to perform secondary screening and filtering of the toilet vehicle attack client according to the distribution state of the updated model parameters of the client relative to the loss value of the WGANGP model; according to the hardware information of the client training model, realizing the third screening and filtering of the toilet vehicle attack client; and aggregating the update models of the remaining clients after three times of screening to obtain a global model, wherein the global model is broadcasted to the block chain for next federation learning training. According to the method, the toilet car attacking client is prevented from participating in aggregation to obtain the aggregation model, defense of federated learning toilet car attacks is achieved, and then privacy of the client and leakage of a global model are protected.

Description

technical field [0001] The invention belongs to the field of federated learning-oriented defense, and in particular relates to a blockchain-based federated learning free-riding attack defense method. Background technique [0002] A common problem in machine learning (ML) is that training a model requires the collection of large amounts of training data, which is often concentrated for analysis. However, when training ML models in a multi-party setting, users must share their potentially sensitive information with the centralized service. Federated learning is a prominent solution for high-scale secure multi-party ML: clients train shared models with a central aggregator without revealing their underlying data or computation. However, doing so poses a subtle threat: customers who were previously passive data contributors are now actively participating in the training process. This brings new privacy and security challenges. [0003] In this case, the models trained by aggr...

AI Technical Summary

Problems solved by technology

Free riders may have two main motivations for submitting false updates: First, the customer may not have the training data and cannot pass the normal training and needs the model to upload the update

Second, the client may not have enough computing power to participate in training or wish to save local CPU and other computing resources

Free riders may try to avoid detection by adding Gaussian noise with zero mean and some standard deviation, such that the resulting gradient update matrix has a similar standard deviation to other clients

Images