Information Processing: Introduction to Artificial Neural Networks

Information Processing: Introduction to Artificial Neural Networks

Information Processing, Volume IV is an incomplete volume. The author passed away suddenly on February 26, 2018, while working on both his Information Processing: Supplemental Exercises for Volume I (published May 30, 2019) and Volume IV of his Information Processing series, entitled 'Introduction to Artificial Neural Networks' (dated June 2017). He had completed about 150 pages of Volume IV. His friends and collaborators, Dr. Romke Bontekoe and Dr. Barrie Stokes, graciously offered to edit the Volume IV draft, update the Mathematica code, and bring the draft into a form ready to be published. They did not attempt to complete missing chapters or add original content. Some chapters listed in the draft table of contents were represented only by placeholders in the main text, and there was no reasonable way to fill in this missing material. The original ‘Contents’ section was amended accordingly. A generic label, artificial neural networks (ANNs), has been bestowed on the whole concept of a biologically inspired structure of interconnected nodes and weights that might mimic in computer code how the human brain solves inferential problems. This volume will provide the interested reader a glimpse into some of the author's ideas concerning ANNs and their application to the general problems of information processing and inference under uncertainty. One common inferential scenario is classification. This volume compares the approach offered by a combination of our preferred Bayesian and MEP attack with the less probabilistically motivated ANNs. It is profitable to compare these approaches since ANNs might offer a way out of the dilemma posed by the dreaded curse of dimensionality. Now that Mathematica is endowed with powerful new machine learning and ANN-related functions, they can be used to tackle the typical inferential problems addressed previously in these books, and extensive numerical experiments with ANNs are possible. For appropriately simplified cases, the Mathematica ANN results are compared to those provided by Bayesian and MEP techniques. The Volume IV Mathematica Notebooks and other supplemental information are available from a public repository on github.com.