Description: Neural networks have fallen only to rise again twice. The latest incarnation borrows much from the past but adds elements to support deep learning. This article demonstrates the state of the art in deep neural network programming.

Description: The author presents autoencoders in a clear manner and then goes on to show an intriguing application where an autoencoder can alert us to events out of the ordinary.

Description: This article discusses how to handle semisupervised data. This type of data might have a large number of observations, yet a small number of labels.

Description: Evaluates the ability of various model types to automatically approximate various mathematical formulas. This can be used as a guide for feature engineering.

Description: This article describes the history and in some respects, the future of Big Data, and the tools we can use to handle it for PM and for machine learning. One such tool is Vowpal Wabbit, which sounds like something from a Bugs Bunny and Elmer Fudd cartoon (“Dwat that wabbit”), but it really is a popular approach to process a dataset of any size, as there is no need to load all the data into memory. Big data is forcing us to enhance many of our tools. So is the increased actuarial usage of stochastic-in-stochastic analyses (nested stochastic processes), which can result in unacceptably long program run times. Many of our popular computer languages are not inherently well suited for parallel computations.

Description: Fuzzy logic is form of multi-valued logic that allows the creation of fuzzy rules. This article introduces the "sets" package for the R programming language for fuzzy logic.

Description: This article discussed the NEAT family of neural network algorithms. NEAT automatically determines the number of hidden neurons, as well as their structure. Additionally HyperNEAT can scale to very large problem domains.

Description: Introduction to PMML and how to use of it in the R programming language. PMML is an interchange format for predictive models.

Description: A discussion on supervised learning--the class of machine learning where a model is trained to produce a specific result for a given input.

Description: Bayesian networks are one of many models that are useful for predictive modeling. This article introduces the concept of Bayesian network and shows how time-series data can be modeled for a Bayesian network.