FedDAN: Federated Learning with Distillation and Adapter Net Model Using in Intrusion Detection System

Intrusion detection technology is one of the core technologies of dynamic security technology. The traffic monitoring method based on machine learning is a more advanced method at present, but this method needs to transmit and share data, which is not conducive to the privacy protection of sensitive data. Federated learning can eliminate data islands and transfer knowledge without exchanging data, which can be used to solve the problem of data privacy. Since intrusion detection technology needs to be deployed on devices with different computing power and data volumes, each client needs to customize the scale and structure of the network. The existing federated learning models are difficult to meet this requirement. In this paper, the FedDAN model is proposed, which uses knowledge distillation technology for knowledge transfer, and the adapter net technology is proposed for the first time to help complete the knowledge transfer between heterogeneous models. Experiments show that the FedDAN model can realize the requirements of model heterogeneity, and can improve the overall recognition accuracy of the model in intrusion detection tasks. FedDAN model solves the data privacy problem in the task of intrusion detection and provides the choice of personalized design model on the premise of ensuring the accuracy of the model, and becomes a new solution in the industry.