Federation Learning of Optimized Convolutional Neural Network Structure for Intrusion Detection

Abstract

While deep learning-based intrusion detection algorithms are a hot topic in intrusion detection research, the majority of research focuses on ways to improve the algorithms' accuracy, overlooking the fact that the limited labeled data generated by a single organization is insufficient to train a deep model with high accuracy in practice. The article discusses an intrusion detection system based on federal learning and convolutional neural networks that may enhance the volume of data by jointly training the model using data sets from several participants. The method uses a federated learning architecture to construct a deep learning-based intrusion detection model. The data dimension is initially reduced to two dimensions via data padding, followed by feature extraction and learning via the federated learning process utilizing the DCNN network, and ultimately, the model is trained with a softmax classifier for detection. Additionally, the method constructed the network structure of the nested Maxout multilayer perceptron layer by optimizing the convolutional neural network model. This enhanced the convolutional neural network's convolutional layer for the purpose of extracting foreground target features and nesting the Maxout multilayer perceptron layer network structure. The experimental results reveal that this method greatly decreases training time while retaining a high detection rate. In addition, the new technology is better at protecting data and personal information than older intrusion detection methods were at doing so.