Intrusion detection method based on joint deep learning

Through federated learning and transfer learning technology, the global model is built in network intrusion detection and model pruning is performed on the client, which solves the problem of insufficient data samples and non-independent and homogeneous data, and improves the accuracy of the model and resource utilization efficiency.

CN119696830AInactive Publication Date: 2025-03-25XIDIAN UNIV
8 Cites 0 Cited by

Patent Information

Application Number
CN202411685894.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-23
Publication Date
2025-03-25
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art faces the problems of insufficient data samples and non-independent and homogeneous data in network intrusion detection, which leads to a degradation of model performance and it is difficult to effectively deploy deep learning models in resource-constrained environments.

Method used

A federated learning-based method is adopted, combining transfer learning and deep learning technology, without sharing private data, a global intrusion detection model is built through collaborative training of the central server and multiple clients, and the model pruning is performed on the client to reduce the demand for computing resources.

Benefits of technology

It improves the accuracy and generalization capabilities of the intrusion detection model, reduces communication costs and computing resource requirements, enhances network security protection capabilities, and effectively deploys deep learning models in resource-constrained environments.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses an intrusion detection method based on joint deep learning, and mainly solves the problems that the number of attack samples of a single mechanism is insufficient, data sharing is unwilling, and a high-accuracy intrusion detection model is difficult to train. The method comprises the implementation steps that a basic model is pre-trained on a public big data set to serve as an initialized global model of a central server, and a client side trains a local data set through feature-based transfer learning so as to adapt to specific data features. Meanwhile, the complexity of a local model is optimized by adopting an L1 unstructured pruning technology in the training process of the client, so that efficient deployment of the model on resource-constrained equipment is realized. The intrusion detection model generated by the method has the advantages of being high in detection rate, strong in generalization ability and suitable for deploying a deep learning model in a resource-constrained environment.
Need to check novelty before this filing date? Find Prior Art