PDF Publication Title:
Text from PDF Page: 015
3.1.1 Machine learning in data networks Machine learning algorithms can learn directly from data without an explicit prior programming task. Machine learning with its ability to learn from data is especially suitable for problems that are too complex to be fully defined or whose definition cannot be made accurately. These are, precisely, the kind of problems that arise in networking. Many machine learning algorithms can be applied to the diverse problems that originate in data networks, such as: Random Forest [11], Gradient Boosting Machine (GBM) [12], Support Vector Machine (SVM) [13], Logistic Regression, Multinomial Logistic Regression [14], Multilayer Perceptron (MLP) [15], K-Nearest Neighbors (KNN) [14], Principal Component Analysis [14], K-Means [14], Naïve Bayes [14], and many more [7]. These algorithms can be subdivided into four main groups: • Supervised: We start with some training data that contains the ground-truth values (responses) that we will then want to predict [14][16]. o Regression: The value to predict is continuous o Classification: The value to predict is discrete/categorical The challenge is to generalize the prediction to test data for which the predicted correct value is not known. • Unsupervised: We start from some training data without values to predict. The goal is to find structure/relationships in the data, without prior information. It can be further subdivided in clustering, dimensionality reduction (latent variables discovery), anomaly/outliers detection, probability density function learning... [14][16]. The challenge is to find the latent structure of the data. • Semi-supervised: We start with some training data that contains the correct responses plus additional training data without them [17]. The objective is similar to the supervised scenario: to predict the correct value for new test data. The challenge is to generalize the prediction to new test data (as in the supervised scenario) while making full use of the additional information provided by the un-labeled data. • Reinforcement learning: The objective is to learn the best sequence of actions that optimize the expected sum of future rewards [18] while interacting with a usually unknown environment that generates the rewards and state transitions that define the dynamics of the interaction. The challenge is to learn a policy function that generates the best current action considering an uncorrelated and possibly sparse sequence of rewards without knowing the dynamics of the environment. An important subset of algorithms employing a cascade of interconnected layers of (usually) non-linear nodes corresponds to the so-called: deep learning algorithms [19]. Deep learning algorithms are used in supervised, unsupervised, semi-supervised and reinforcement learning models. They are usually extremely good in representation learning which is a real advantage to avoid difficult and costly feature engineering. The training is usually done by optimizing a cost function using some form of gradient descent. The main deep learning algorithms considered in this thesis are: Convolutional Neural Networks (CNN) [20], Recurrent Neural Network (RNN)[21], Long Short-Term Memory Doctoral Thesis: Novel applications of Machine Learning to NTAP - 13PDF 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 (Standard Web Page)