IMPROVEMENT OF THE USE OF THE PRIVATE MODEL BY MINIMUM OF THE EXPECTED DEVELOPMENT

摘要

Training a model is performed to minimise expected noise loss (ELUN) while maintaining differential privacy.Noise is added to the weights of a machine learning model as random samples taken from a smoking authorisation, adding noise in accordance with a private sphere budget.The ELUN is minimized using a loss function that anticipates the noise added to the weights of the machine learning model to find a point in the parameter space,for which the loss is robust against the noise in the weights.Adding noise and minimizing the ELUN are iterated until the weights converge and optimization limitations are met.The model is used for any input, while the privacy of the training data used to train the model is protected.