New Bulgarian University > Center for Cognitive Science > Preparatory Program > Course Description

COG506 Machine Learning
  1. Aims:
    On completion of this course the students should have acquired the fundamental ideas, algorithms and techniques of Symbolic Machine Learning (excluding Neural Network Learning and Genetic algorithms), i.e. the Artificial Intelligence approach to ML. They should be able to read and understand specialised ML literature, use existing ML systems and implement simple ML algorithms.

  2. Learning strategies:
    Tutoring and computer exercises with ML algorithms and systems.

  3. Overall duration and format:
    One semester (15 weeks) with 2 hours seminars and 1 our lab work per week.

  4. Credit hours: 2

  5. Lecturer: Zdravko Markov - Assoc. Prof., Institute of Information Technologies - Bulgarian Academy of Sciences

  6. Literature:
    Markov, Z.,

    Gennari, J.H., Langley, P., Fisher, D.,

    Genesereth, M. and N. Nilsson,

    Luger,G.F. and W.A. Stubblefield,

    Michalski, R. and R. Stepp.,

    Michalski, R., J. Carbonell and T. Mitchell,

    Mitchell, T.,

    Mitchell, T. M., Keller, R. M. and Kedar-Cabelli, S. T.,

    Mitchell, T. M., Utgoff, P. E. and Banerji, R.,

    Muggleton, S.,

    Muggleton, S. (ed.),

    Quinlan, J.R.,

    Quinlan, J.R.,


  7. Course outline:
    The course gives a broad overview of the ML approaches and further focuses on some of them mostly within the Inductive paradigm. Some basic algorithms for Unsupervised learning are also discussed. The inductive learning approaches are emphasised since they are natural extensions and applications of classical AI techniques.

  8. Main Topics:

    The material is organised in 7 topics each one discussing a typical ML problem, the basic techniques used for its solution and an well-known algorithm or system related to it. The topics are accompanied with a number of exercises and a selected bibliography. They are as follows:

    Topic 1: Basic concepts, Learning semantic networks. The basic paradigms, strategies and techniques of ML are introduced. The problem of learning semantic network descriptions from examples is discussed in detail, further illustrating the basic concepts of inductive learning.
    Required reading:


    Topic 2: Inductive Learning, Version space. The induction problem is defined formally and further illustrated by describing the classical candidate elimination algorithm for version space search. A simple attribute-value language which is used in the following topics is introduces. Required reading:


    Topic 3: Induction of Decision Trees. The decision trees as a hypothesis language are introduced. The basic algorithm from the ID3 family is described and illustrated with examples.
    Required reading:


    Topic 4: Conceptual clustering. The basic ideas of unsupervised learning are presented. Two well-know algorithms for conceptual clustering are discussed - CLUSTER/2 and COBWEB.
    Required reading:


    Topic 5: Explanation-based learning (EBL). The basic ideas of learning in a presence of a domain theory are presented. The advantages and drawbacks of EBL are discussed in connection with the traditional inductive (similarity based) approaches
    Required reading:


    Topic 6: Occam's Razor, Bayes Learning and Minimal Description Length Principle. The basic ideas and techniques within the framework of the complexity-based approaches to ML are presented.
    Required reading:


    Topic 7: Inductive Logic Programming (ILP). The recently emerged area of Inductive Logic Programming is discussed in the framework of the classical induction problem. The basic ILP techniques, such as clause subsumption, inverse resolution and predicate invention are described. The ILP strategies are illustrated with examples.
    Required reading:


  9. Assessment:
    Based on answering questions and solving illustrative problems.

  10. Prerequisites:
    The material is based on some fundamental notions of First Order Logic and Logic Programming (Predicate calculus, Resolution etc.) which are common for any AI course. The exercises are based mainly on Prolog. Therefore a general AI course and a Prolog programming course are desirable to be taken before attending the ML course.

New Bulgarian University Center for Cognitive Science

[ Department ][ Center ][ Research ][ Programs ][ Events ][ Faculty ][ Staff ][ Students ][ Mail ]

Last updated on 2000.01.31, 08:00; please report errors to webadmin