Course syllabus

# Artificiella neuronnätverk

Artificial Neural Networks

## EXTP80, 7,5 credits, A (Second Cycle)

## General Information

## Aim

## Learning outcomes

## Contents

## Examination details

## Admission

Admission requirements:
## Reading list

## Contact and other information

Artificial Neural Networks

Valid for: 2016/17

Decided by: Education Board B

Date of Decision: 2016-03-29

Language of instruction: The course will be given in English on demand

The aim of the course is to give the student knowledge about artificial neural networks, both theoretical knowledge and how to use them in practical applications such as pattern recognition, function approximation and optimization problems.

Knowledge and understanding

For a passing grade the student must

- understand and be able to use the simple perceptron for linear problems.
- understand and be able to use the multilayer perceptron, including methods for learning, choices of error function, model selection and generalization.
- understand the meaning of feedback networks and its usage within time series analysis.
- understand and be able to use various types of feedback networks, such as the FIR-network, networks with time lags, multilayer perceptrons with feedback and networks with context units.
- understand and be able to use networks for principal component analysis, networks for clustering and networks for supervised learning vector quantization (LVQ).
- understand and be able to use Self-organizing feature maps (SOFM).
- understand and be able to formulate simple combinatorial optimization problems and use feedback networks to find approximate solutions to such problems.
- understand and be able to use the mean field approximation in connection with networks for combinatorial optimization.

Competences and skills

For a passing grade the student must

- be able to describe and handle network ensembles and describe Bayesian training of multilayer perceptrons.
- be able to describe and handle fully connected feedback networks for associative memories (the Hopfield model) and the simulated annealing optimization technique.
- be able to write a computer program that trains a multilayer perceptron for a binary classification problem and be able to evaluate the performance of the network.
- bea ble to show why an ensemble of networks often performs better compared to a single network.
- be able to write a computer program that uses the Hopfield
model to find approximate

solutions to the graph bisection problem.

- Feedforward networks
- Feedback networks
- Self-organizing networks
- Networks for combinatorial optimization

Grading scale: TH

Assessment: To pass the entire course, a passed test, passed reports on the computer exercises and participation in all compulsory course elements is required.

The number of participants is limited to: No

- As posted on our webpage and billboard.

Course coordinator: Mattias Ohlsson, mattias@thep.lu.se

Course homepage: http://cbbp.thep.lu.se/~mattias/teaching/fytn06/