Differential privacy budget optimization method based on deep learning in Internet of things environment

Abstract

In order to effectively deal with the massive data brought by the large-scale application of the Internet of things , Deep learning is widely used in the Internet of things . However , Depth model in the training process , Existential reasoning attack 、 Model reverse attack and other security threats , This will lead to the disclosure of raw data in the input model . Differential privacy is applied to protect the parameters of the depth model training process , It is an effective way to solve this problem . Based on this, a differential privacy budget optimization method based on deep learning in the Internet of things environment is proposed , According to the iterative change law of parameters , Allocate different budgets adaptively ; To avoid the problem of excessive noise , The regularization term is introduced to constrain the perturbation term , It can prevent the neural network from over fitting , It also helps to learn the salient features of the model . Experiments show that , The proposed method can effectively enhance the generalization ability of the model ; As the number of iterations of the model increases , The model after noise training , And the model trained with raw data , The accuracy difference between the two is less than 0.5%. therefore , The proposed method can realize user privacy protection , At the same time, the availability of the model is effectively guaranteed , Achieve a balance between privacy and availability .

key word ：