Financial Risk Control Scheme Based On Neural Network And Security Gradient Mean Value

The risk control of personal and corporate loans has always been a major problem for Banks and lenders.The increasing development of IT technology offers hope for a solution to this problem.In the past,centralized machine learning was used to assess the risk of users.However,the data itself exists in a decentralized form,and some data holders are unwilling to share the data directly.Moreover,in recent years,laws related to privacy protection have restricted the data used by enterprises and individuals.How to evaluate financial risk safely and efficiently has become an urgent problem in this field.In view of the above problems,this paper adopts the method of neural network and safety federal mean to establish a financial risk assessment scheme,so as to evaluate the unmarked customers.Compared with the existing work,this scheme has the following advantages:(1)There is no need for centralized learning,each organization can train the model locally and hide the gradient.(2)When there are no dropped users,the number of operation rounds is less than the existing protocol.(3)For the case of dropped lines and delay,there is a corresponding key recovery mechanism.(4)The accuracy rate is higher than that of each user.(5)It is more' efficient than homomorphic encryption.Finally,the security and efficiency of the scheme are analyzed,and the scheme is implemented to verify the security and efficiency of the scheme.In this paper,the neural network and security gradient mean are combined for the first time and applied in the field of financial risk assessment.The neural network is used to realize the risk assessment with high accuracy.The security gradient mean is used to solve the problem that centralized machine learning and deep learning cannot guarantee user privacy.The proposal of this scheme has high application value and practical significance in the field of financial risk assessment.