M330: Neural Networks

Course Objective This two-day course enables participants to develop their own neural network solutions using Mathematica and the Neural Networks application package and gives attendees the relevant background theory to understand the applicability and limitations of neural networks.

Course Summary This course presents the theory and practice of neural networks with Mathematica and focuses on the Neural Networks package. The features and capabilities of the package are demonstrated, and numerous examples and practical hands-on exercises are included. The material is presented as a sequence of five lectures, each one followed by a problem session to help attendees understand the material and to provide a focused and practical learning experience. The lectures cover the different kinds of neural networks and the types of problems for which neural networks are used. Basic theoretical concepts, illustrated with graphs, figures, and examples, are covered to support practical neural network training.

Presenter The course is presented by Jonas Sjöberg, the developer of the Neural Networks package, or another Wolfram Education Group certified instructor. Professor Sjöberg has 10 years of experience in teaching neural networks in industry and academia.

Target Audience The course is designed primarily for people who want and need to estimate relations in data using Mathematica. Attendees typically have wide-ranging backgrounds and include engineers and professionals who work with all kinds of data, including technical, medical, and economic data.

Delivery Type Courses are delivered as instructor-led classes in computer classroom facilities or as online classes over the web. Course topics are presented with alternating sessions of lectures and exercises. All classes feature low student-teacher ratios.

Syllabus This basic course is organized into five segments.

Introduction
Overview of neural network history and types of problems: function approximation, classification, data clustering, time series, and dynamic systems
Feedforward Neural Networks and Radial Basis Function
Learning, overlearning, and initialization of neural networks
Theory and Background of Neural Networks
Description of the inherited problems when functions are fitted to data, possibilities for handling these problems using neural networks, and practical aspects
Nonlinear Dynamic Black-Box Modeling
Modeling of time series and dynamic systems using linear and nonlinear models
Classification and Clustering with Neural Networks
Two classes, many classes, neural network classifiers and relations to other classifiers, the perceptron as classifier, nearest-neighbor classification, vector quantization, unsupervised methods, self-organizing maps, and the Hopfield network

Course Materials Each attendee will be provided with Mathematica course notebooks and access to the current version of Mathematica. The course notebooks require Mathematica or Mathematica Player. For attendees participating in classroom-based sessions, course materials are distributed in print and on CD-ROM, and are yours to keep; a computer running Mathematica is available for your use during class. For attendees participating in online classes, a download of the course materials is provided; a temporary Mathematica training license is provided upon request.

Prerequisites Course attendees are expected to have basic familiarity with Mathematica approximately equivalent to that provided by "M101: A First Course in Mathematica." A basic-level knowledge of signal/image processing concepts and experience with introductory computer programming are also helpful.