Deep learning model fairness improvement system and method based on adversarial disturbance

Abstract

The invention discloses a deep learning model fairness improvement system and method based on adversarial disturbance, the system comprises a deployment model, a disturbance generator and a discriminator, the deployment model comprises a feature extractor and a label predictor, and the disturbance generator is connected with the feature extractor. A deployment model is prevented from extracting sensitive features of data, and fairness is improved under the condition that the model is not changed. The input data of the deployment model is processed, and the deep learning model does not need to be changed. According to the method, model fairness is improved based on antagonistic disturbance, a corresponding disturbance generator and a discriminator are designed, sensitive attribute information related to fairness is captured by using the discriminator, training optimization of the disturbance generator is guided, sensitive attribute information of antagonistic disturbance hidden data is generated, and information related to a target task is reserved; and the model is prevented from extracting sensitive information of input data in a feature extraction process, so that the prediction fairness is improved.

Description

technical field [0001] The invention relates to the field of trusted artificial intelligence (AI), in particular to a system and method for improving the fairness of a deep learning model based on an adversarial disturbance. Background technique [0002] In recent years, deep neural networks have demonstrated excellent performance in many fields, such as image processing, natural language processing, speech recognition, and so on. Although the popularization of artificial intelligence technology has promoted changes in various fields and brought convenience and improvement to human life, studies have found that some existing artificial intelligence systems have ethical risks. They contain prejudice and discrimination against specific groups, and even disadvantaged groups put in a more disadvantaged position. Therefore, alleviating the bias of deep learning models and improving the fairness of model decision-making are important prerequisites for ensuring the reliable applic...

AI Technical Summary

Problems solved by technology

Existing techniques to improve the fairness of deep learning models essentially need to modify the deployed model to prevent the model from learning false associations to eliminate prejudice against specific groups, thus greatly limiting the practical application of these mechanisms for improving model fairness

Images