Explore chapters and articles related to this topic
Signature Generation Algorithms for Polymorphic Worms
Published in Mohssen Mohammed, Al-Sakib Khan Pathan, Automatic Defense Against Zero-day Polymorphic Worms in Communication Networks, 2016
Mohssen Mohammed, Al-Sakib Khan Pathan
In the previous section, we discussed the supervised ML techniques. In this section, we give an overview of unsupervised learning from the perspective of statistical modeling. Unsupervised learning can be motivated from information theoretic and Bayesian principles. We briefly review basic models in unsupervised learning, including factor analysis (FA), principal components analysis (PCA), mixtures of Gaussian models, independent component analysis (ICA), hidden Markov models (HMMs) [35], state-space models (SSMs), and many variants and extensions. We derive the expectation-maximization (EM) algorithm and give an overview of fundamental concepts in graphical models and inference algorithms on graphs. Then, we present a quick review of approximate Bayesian inference, including Markov chain Monte Carlo (MCMC), Laplace approximation, Bayesian information Criterion (BIC), variational approximations, and expectation propagation (EP). We assume that the reader is familiar with elementary linear algebra, probability theory, and calculus but not much else. Before starting the discussion of unsupervised learning models, we present a brief introduction about the following: Unsupervised learningMachine learning, statistics, and information theoryBayes rule
Bayesian estimation of discrete choice models: a comparative analysis using effective sample size
Published in Transportation Letters, 2022
Jason Hawkins, Khandker Nurul Habib
In addition to the methods explored in this paper, many advances are being made that are likely of interest to the transportation researcher. Other methods of approximate inference have been developed, such as operator variational inference (Ranganath et al. 2016) and expectation propagation (Vehtari et al. 2020). Expectation propagation is useful when dealing with large datasets because it partitions the problem into a set of local inferences rather than focusing on the global posterior. These algorithms, like ADVI, are useful tools for applying Bayesian methods to large datasets but remain relatively experimental – i.e., lacking in documentation and diagnostic tests. Other computational methods developed for machine learning are also seeing applications in Bayesian inference. TensorFlow is based on the representation of a model as a series of tensors forming a computational graph. This structuring of the model allows TensorFlow to run models on GPU and TPU, as well as employ emerging methods such as accelerated linear algebra (XLA) (TensorFlow developers 2020). These tools provide opportunities for transportation researchers to bridge the gap between classical statistics and machine learning through the development of Bayesian neural networks and other methods. However, they do require more advanced computing resources and coding expertise beyond that required for implementation of similar models in Stan.
Federated data analytics: A study on linear models
Published in IISE Transactions, 2022
Xubo Yue, Raed Al Kontar, Ana María Estrada Gómez
To resolve the aforementioned issues, we resort to Expectation Propagation (EP) (Minka, 2001) to approximate the posterior distribution. EP is one of the most widely-used algorithms for computing an approximate posterior distribution (Minka, 2013; Vehtari et al., 2020). Here, we first briefly introduce the idea of EP in a centralized regime. Consider a posterior distribution with independent data points