PDF Publication Title:
Text from PDF Page: 016
Network (LSTM) [22], Variational Autoencoders (VAE) [23] and Conditional Variational Autoencoders (C-VAE) [24][25]. The following section (section 3.1.2) is dedicated to presenting these models in more detail. This thesis is mainly focused in supervised algorithms for classification and precisely in: • Classic ML: o Random Forest, GBM, SVM, Multinomial Logistic Regression, Gaussian Process and MLP. • Deep Learning: o CNN, LSTM and combinations of both. o Generative models: VAE and variants. In addition to the machine learning models mentioned above, and belonging to the same family of predictive algorithms, we have considered some algorithms that are normally used to make predictions for time-series data. These algorithms are not usually studied as machine learning methods, but rather in the area of statistics and time-series analysis. These algorithms have been the usual resort to make predictions when dealing with time-series data, and we have employed them as part of the methods analyzed to predict future traffic in IoT networks. The time-series algorithms considered are: Hidden Markov Model (HMM) [26], Exponential Smoothing [27], AutoRegressive Integrated Moving-Average (ARIMA) [28] and ARIMA with eXogenous covariates (ARIMAX) [28]. The authors in [6][8]provide two excellent and updated reviews on the application of machine learning in networking. In [6] the application areas of ML to networking are structured as: information cognition (route measurement), traffic prediction, traffic classification, resource management, network adaption (QoE optimization, congestion control, routing strategy), performance prediction (QoE and throughput prediction) and configuration extrapolation (automated network protocol and architecture design). The authors consider that all areas will be impacted by ML with a greater future impact on: resource management, network adaption and automated architecture design. In [8] similar (but not identical) application areas of ML to networking are identified: traffic prediction, traffic classification, traffic routing, congestion control, resource management, fault management, QoS/QoE management and network security. Traffic prediction is mainly dominated by time-series algorithms (ARIMA, HMM...) with some few applications of non- time series methods based on: MLP, CNN/RNN and SVM. Traffic classification has employed supervised (K-NN, SVM, Adaboost, XGBoost, Decision Trees...) and unsupervised (K-Means, DBSCAN) methods with no reported deep learning works in the study. Traffic routing is driven by reinforcement learning algorithms, mainly Deep Q-Learning. Congestion control is managed with packet loss classification algorithms using classic supervised models (SVM, K-NN, MLP...) and Queue management is the territory of queue length predictors (ordinary linear regression and time-series models) and adaptive mechanisms based on reinforcement learning plus control theory. ML for Resource management is mainly focused on: a) admission control and b) resource allocation; applying prediction and decision-making algorithms (MLP and Q-learning). Fault management is broken down into: predicting the fault, detecting the fault, localizing the root cause of the fault and automated mitigation of the fault; Doctoral Thesis: Novel applications of Machine Learning to NTAP - 14PDF 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 |