ITI43210 Machine Learning (Spring 2015)

Machine learning is about computers learning through training and experience instead of being explicitely programmed for a given task.The students will get acquainted with several methods and algorithms for machine learning. Based on this, the students should be able to select the methods best suited for the problem in question.

The course should give the students knowledge about the basic properties common to all machine learning methods. Examples include ability to generalise and heuristic search.

The course contains three projects, one about decision trees, rules and regression analysis, one about neural networks, and one about evolutionary computation.

Induction of decision trees and some applications such as medical diagnosis and credit evaluation.

Artificial neural nets and optimization algorithms such as steepest descent and trust region Newton methods. Applications of neural nets to sound and image analysis.

Basic theory for machine learning, for example Bayes' formula, maximum likelihood and the minimum description length principle.

Instance based learning such as nearest neighbour, locally weighted regression, and radial basis functions.

Evolutionary computation, especially genetic algorithms and genetic programming. General principles for evolution. Selection methods and genetic operators such as mutation and crossover. The Baldwin effect.

Some of the topics above require basic knowledge of statistics and information theory which will be taught as needed.

Knowledge

The candidate shall have

• an overview over the most important methods in machine learning, and have a deeper knowledge of three of them, viz. Decision Trees, Neural Networks, and Evolutionary Computation
• knowledge about the practical challenges in connection with data mining, e.g. overfitting, missing values, and classifications with different costs.
• gained knowledge in basic topics such as numerical optimisation and statistical methods, for example Bayesian methods used in search engines like Google and in spam filters.

Skills

The candidate is able to

• arrange or code data to fit data mining and machine learning algorithms
• choose correct tools for a given type of data
• decide on how good the results are based on simpel statistical analysis of, for instance, classification exactness
• use machine learning i practical applications and be able to transfer machine learning models to programming language code

General competence

The candidate has improved his/her competence in

• research and development, for instance finding relevant literature and understand scientific articles about machine learning
• writing scientific texts in English
• treat and analyse data of arbitrary type, even if this is done without inductive learning
  

Portfolio and individual take home exam.
The exam consists of both a portfolio and a individual take home exam.

The portfolio (determines 65% of the final grade) consists of:

• one project on decision trees
• one project about neural nets
• one project in automatic programming.

The projects can be carried out individually or in groups of two students. The students will get an individual grade.

There is also a three day individual take home exam that determines 35% of the final grade.

The student will get an individual joint grade for the entire course. Grades: A - F.

Upon re-examination, both parts of the examination must be retaken. Upon re-examination, new assignments and the take home exam will be decided by the course instructor.

This course is evaluated by a

• Mid-term evaluation (voluntary)
• End evaluation (compulsory)

The lecturer compiles a report based on the evaluation forms filled in by the students and his/her own experience with the course. The report is the discussed by the study quality committee of the faculty of Computer Sciences.