A Survey On Anomaly Detection Model Based On Deep Learning

The main purpose of anomaly detection is to detect sparse data that deviates from the characteristics of the other data,which has important research value in many practical applications.With the rapid development of big data,the dimension and quantity of data have been greatly improved,the traditional anomaly detection model can not meet the needs of traditional anomaly detection,so the anomaly detection model based on deep learning has attracted more and more attention.Hybrid model and generative model are the main models of deep anomaly detection.Hybrid model combines with feature extraction algorithm and traditional anomaly detection algorithm,so that the efficiency is greatly improved.The generative model can generate the training data and has powerful representation learning ability,so it is widely used in anomaly detection in high-dimensional data space.The advantages and disadvantages of the existing deep anomaly detection model are discussed in detail,and two improved deep anomaly detection models are proposed.The main tasks of this paper are:(1)Discussion and analysis of the work related to deep anomaly detectionThis paper introduces the research status of deep learning.The anomaly detection technology based on reconstruction error and hybrid model are combed in detail,and the advantages and disadvantages of the two models are analyzed.According to the latest literature,the relevant techniques of deep learning in the field of anomaly detection are summarized.(2)Anomaly detection model based on multi-grained scanning and AutoencodersFirstly,the isolation mechanism of random hyperplane is proposed to improve the ability of detecting local anomaly from complex data pattern in isolation forest.Secondly,a multigrained scanning is proposed for feature selection,and the feature extraction is carried out by using Autoencoders,so that two hybrid models are constructed.In the experiment with dataset such as KDD99 and P53 Mutant,the structure of the hybrid model improves the stability and performance.(3)Anomaly detection model based on optimization Adversarial AutoencodersFirstly,the problem of Adversarial Autoencoders for anomaly detection is described.Secondly,the evaluation network is used to optimize the Adversarial Autoencoders.Finally the anomaly detection experiment is carried out in the picture dataset MNIST and CIFAR-10,and the comparison experiment is made with the hybrid model.Lastly,summary the paper and prospect the future.