Neural Networks (FFR 135)

Statistics, Optimisation, and Learning

(7.5 credit units)

Bernhard Mehlig (lectures, examiner) Office: S3050
Marina Rafajlovic (examples classes)

Course evaluation

Students participating in the course evaluation:
Tom
Christian

Schedule

Office hours (Marina Rafajlovic)

Fridays 14:00 to 15:00 in S3047

Abstract

This course is intended to describe the use of neural network models in learning and optimisation, e.g., pattern recognition, routing, and prediction. This course is divided into two parts. The first part (3 weeks) provides an introduction to neural networks, focussing on the so-called Hopfield model, its statistical mechanis and optimisation algorithms. The second part (4 weeks) provides a more detailed introduction to learning, describing models, algorithms, and applications.

Table of contents

• An introduction to neural networks: neuroscience, statistics and optimisation
  • Neuroscience
  • An example of a neural network: the Hopfield model
  • Statistics and Optimisation (Monte-Carlo methods, simulated annealing)
• Learning: models, algorithms, and applications
  • Supervised learning: simple perceptrons and layered networks
  • Performance of multilayer perceptrons
  • Unsupervised learning
  • Reinforcement learning
  • Recurrent networks and time series analysis

Literature

1. S. Haykin, Neural Networks: a comprehensive foundation, 2nd ed., Prentice Hall, New Jersey (1999)
2. J. Hertz, A. Krogh, and R. G. Palmer, Introduction to the theory of neural computation, Addison-Wesely, Redwood City (1991). This book is out of print. Copies are available at Chalmers library and in Bernhard Mehlig´s office.

Examination

Credits for this course are obtained by solving the homework set (solutions of examples and programming projects). There will be five sets of homework which are graded.

Every student must hand in her/his own solution on paper. Same rules as for written exams apply: it is not allowed to copy any material from anywhere unless appropriate reference is given. All figures must have axis labels and captions giving all information necessary to reproduce the figure. Describe your results in words. Always compare with theory. Summarise problems, discuss possible reasons. Program code must be appended.

Each of the five examples sheet gives 5 points. In order to pass the course at least 14 points are required.

The examples sheets will be processed by URKUND. Your solutions should be submitted before the deadline

• as PDF files electronically to Bernhard.Mehlig.chalmers@analys.urkund.se and
• as a hardcopy in the usual way (to Marina Rafajlovic)

If you have written your solutions by hand it is sufficient to submit in the usual way to Marina Rafajlovic.

Information about the URKUND system can be found here.

Examples

• Sheet 1 [pdf] Tuesday September 13 at 12:00
• Sheet 2 [pdf] Tuesday September 20 at 17:00
• Sheet 3 [pdf,training set,validation set,explanation] Tuesday October 4 at 17:00
• Sheet 4 [pdf,data,explanation] Tuesday October 11 17:00
• Sheet 5 [html] Tuesday October 18 at 17:00

Updated: Monday, October 10, 2011 4:16:29 PM by Bernhard Mehlig