Research on intrusion detection method based on knowledge distillation under unbalanced data set
The network intrusion detection model based on deep learning is faced with the problems of complex mod-el structure,low deployment efficiency and unbalanced traffic data categories.To solve these problems,a network in-trusion detection method combining knowledge distillation and class-weight focus loss is proposed.In this method,the intrusion detection model with high precision and large number of parameters is used as the teacher model to generate distillation loss with the small student model.The focus loss function with increasing category weight is in-troduced as student loss.The total loss function is generated by combining distillation loss and student loss to opti-mize the student model.The experimental results show that the method has some improvement in each index com-pared with the non-distillation model.
intrusion detectiondeep learningknowledge distillationunbalanced datafocal loss
万方数据
维普
