Health crowd sensing system and cost-optimized federated learning method thereof

Abstract

The invention discloses a health crowd sensing system and a cost-optimized federated learning method thereof, and provides a system architecture of a cloud edge end and a cost-optimized federated learning model, and the federated learning is deployed in a cloud edge cooperation mode to train a global model; and in the cost-optimized federated learning model, whether to participate in this round of training is determined according to verification precision of a to-be-trained global model, so that communication cost and local training cost are optimized, the cloud performs quality evaluation on received local model update and selects high-quality local model update to participate in aggregation of the global model, and the learning efficiency is improved. A large number of experiments based on a public data set prove that the method provided by the invention effectively reduces the communication cost and the training cost on the premise of ensuring the global model precision.

Description

Technical field [0001] The invention belongs to the field of health and medical technology, in particular to a group of Chilean health system and its perception federal learning method cost optimization. Background technique [0002] Learning is under federal privacy data hold multi-party co-training machine learning paradigm, its goal is to train high-quality global model. It acquires a local machine learning model by iteratively updating the model and a polymerization global model parameters, the learning algorithm in each round in this paradigm, the client holds the data independently of each model according to its local training data and the model parameters are transmitted to the central coordinator, the coordinator of the central partial summary model parameters client and calculating a new global model. The learning algorithm is repeated until the global model to meet certain convergence criteria. Federal learn in some applications have been successful, such as the one at ...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem of high model training cost and high parameter transmission cost in the federated learning process in the existing healthy crowdsensing scene, the present invention proposes a healthy crowdsensing system and its cost-optimized federated learning method

Images