Article Preview
Top1. Introduction
With the development of new Internet application technologies such as Internet of things, cloud computing and big data, cyberspace security is becoming more and more important. Among them, how to effectively identify network attacks is the core issue of cyberspace security. Intrusion detection technology is an important part of cyberspace security. It can detect various intrusion data through collection and analysis of various information on the network (Yongzhong & Rushan 2010), which is the focus of research in the field of computer network security (Ding & Li 2012). The commonly used detection techniques for intrusion detection are misuse detection and anomaly detection (Zhang & Yong-Zhong 2013). Among them, anomaly detection has attracted much attention because it can detect new types of intrusion behavior, and has become a hot spot in intrusion detection research. With the advent of the era of big data, the methods of network intrusion is also being updated. The new network intrusion presents a trend of intelligence and complexity. Traditional anomaly detection technology is difficult to achieve the desired effect for these new types of network intrusion. Faced with a variety of new network intrusions, a variety of new intelligent intrusion detection algorithms have emerged. Deep learning is an important branch of machine learning. Its structure is to simulate the human brain neural network (Le 2013). After learning the underlying raw data, it gets more abstract high-level features. Deep learning uses a layer-by-layer greedy algorithm and uses a combination of classifiers such as SVM and Logiestc to classify. It has been widely used in image processing (Li-Cheng 2012), speech recognition, natural language processing and other aspects (Yu et al., 2015), which is a kind of intelligent learning. Emerging areas. Deep learning is a new field of intelligent learning, which uses the combination of layer by layer greedy algorithm, SVM, logiestc and other classifiers to classify. It has been widely used in image processing (Li-Cheng 2013), speech recognition, natural language processing and other aspects (Yu et al., 2015). The features extracted by deep learning are highly complete, so this paper combines autoencoders and softmax and applies them to intrusion detection. In addition, semi supervised learning is used to train with a small amount of labeled data and a large amount of unlabeled data, which reduces the characteristics of sample requirements. Specific to the problem of low accuracy in intrusion detection system, this work proposes an intrusion detection algorithm based on semi-supervised learning and deep learning. In this paper, deep learning is used to extract features with high feature completeness (Schulz & Behnke 2012; Chun-Lin et al., 2014), and the sparse autoencoder and softmax classifier are combined to apply to the intrusion detection algorithm. In addition, this paper also uses semi-supervised learning to train with a small amount of labeled data and a large amount of unlabeled data, which reduces the sample requirements. Aiming at the problem of low accuracy in intrusion detection system, this work proposes an intrusion detection algorithm based on semi supervised learning and deep learning. Experimental studies began by using intrusion detection evaluation data, namely KDD CUP99, to benchmark the performance of the proposed approach. Experimental results show that the algorithm effectively improves the detection rate and reduces the false positive rate.