PDF Publication Title:
Text from PDF Page: 078
7.5 Paper 5: Variational data generative model for intrusion detection Manuel Lopez-Martin, Belen Carro, Antonio Sanchez-Esguevillas Variational data generative model for intrusion detection Knowledge and Information Systems 2.247 Q2 0 (https://scholar.google.es/citations?user=3RSZbOYAAAAJ&hl=es) Published: 13-December-2018 https://doi.org/10.1007/s10115-018-1306-7 7.5.1 Objectives In order to train an intrusion detection classifier is very important to have access to representative and balanced training data. This is usually a difficult task since intrusion detection samples of network traffic are strongly biased to normal traffic, being difficult to access traffic associated with intrusion events. Considering these difficulties, it is important to have a way to produce traffic samples associated to intrusion events which are rare compared with the main/normal traffic. There are several classic techniques to over-sample the minority classes in order to have a more balanced dataset (e.g. SMOTE, ADASYN...). These techniques create new data points based in the proximity to existing points of the same class. They are based on topological proximity and do not consider the probability distribution of features for the different classes. In this work is presented a new method to create synthetic data based on their probability distribution conditioned on the class to which they belong. This new method consists of a generative model based on a customized Variational Autoencoder (VAE). The VAE architecture has been modified to create a novel model based on a conditional VAE which integrates the class label as a new input to the VAE’s decoder architecture. The advantage of using a conditional generative model to generate new data is based on the capacity of this model to create data using noise as input, therefore we do not create synthetic samples based on proximity to existing samples (a noisy and error prone task) but on following a probability distribution for the minority class which is learned as part of the training of the resulting model. Besides the above-mentioned advantage as an easier and more principled way to generate synthetic data, we show that the novel architecture, based on the conditional VAE, produces synthetic data that provides better results when this data is used as additional data to train a Authors Title Journal Impact Factor Quartile #Citations Status Link Doctoral Thesis: Novel applications of Machine Learning to NTAP - 76PDF Image | Novel applications of Machine Learning to Network Traffic Analysis
PDF Search Title:
Novel applications of Machine Learning to Network Traffic AnalysisOriginal File Name Searched:
456453_1175348.pdfDIY PDF Search: Google It | Yahoo | Bing
Cruise Ship Reviews | Luxury Resort | Jet | Yacht | and Travel Tech More Info
Cruising Review Topics and Articles More Info
Software based on Filemaker for the travel industry More Info
The Burgenstock Resort: Reviews on CruisingReview website... More Info
Resort Reviews: World Class resorts... More Info
The Riffelalp Resort: Reviews on CruisingReview website... More Info
| CONTACT TEL: 608-238-6001 Email: greg@cruisingreview.com | RSS | AMP |