Artificial Intelligence 1

The course is a substitute for:

Artificial Intelligence 1 (E33UI1)
Artificial Intelligence 1 (X33UI1)

Lecturer:

Tutor:

Supervisor:

Department of Cybernetics

Synopsis:

The course covers main topics of symbolic AI, i.e. those closely bound to knowledge representation in logic and automated theorem proving and its utilization in declarative programming, namely Prolog - its advaitages are demonstrated on specific solutions of typical general AI problems (state-space search, methods of common sense reasoning). Machine learning and uncertain data. Finally, inductive logic programming is explained as a new perspective method extending significantly application possibilities of machine learning.

Detailed up-to-date information and lecture slides are available on the server datel.felk.cvut.cz

Requirements:

During the labs, each student is supposed to accomplish 2 unaided tasks rated 10 points each (provided the solution is submitted in time). Perfectly solved written exam corresponds to 40 points. The minimal pass condition for the course is to gain at least 50% of points both in labs and for the written exam.

Syllabus of lectures:

1. Knowledge representation and 1. order logics, its language and methods

2. Automation of theorem proving. Refutation by resolution and its properties

3. Automated theorem provers, declarative a logic programming, Prolog language

4. Wrap-up: logics in AI, its advantages and limits. Utilization for problem solving. Problems of complexity

5. Common sense reasoning, qualitative simulation and reasoning with uncertainty

6. Inductive machine learning - basic notions, learning as a search

7. Review of symbolic methods I.: building decision and regression trees

8. Review of symbolic methods II.: rule based systems - AQ, CN2, association rules

9. Reinforcement learning - motivation and methods: dynamic programming, Monte Carlo, TD learning

10. Inductive logic programming (ILP) - basic idea, background knowledge and its role in ILP

11. Principles of ILP systems - training examples and their properties Practical ILP applications

12. Review of formalisms for uncertainty handling: conditional probability and its properties, Dempster-Shaffer theory, fuzzy logic, non-monotonic reasoning. Knowledge representation and Bayes nets

13. Independence in a Bayes net, design of its structure and estimation of its parametres. EM algorithm. Probabilistic modelling of multirelational data

14. Computational learning theory, PAC

Syllabus of tutorials:

1. Transformation of a problem from natural language into propositional calculus

2.-3. Resolution as a tool for problem solving

4.-5. Some existing systems for automated theorem proving (Otter, Prolog, CLP, ... )

6. Test

7. Some existing systems for machine learning: WEKA, etc

8. Preprocessing data and machine learning. SumatraTT

9. Individual work: solving assigned tasks

10. Existing ILP systems, e.g. Progol

11. Individual work: solving assigned tasks

12. Evidence based reasoning (bayesian classification). Bayes nets in fuzzy version of FEL expert

13. Individual work: solving assigned tasks

14. Final evaluation

Study Objective:

Study materials:

Selected chapters from the following books:

[1] Russell, S., Norvig, P.: Artificial Intelligence, A Modern Approach, Prentice Hall Series in AI, Englewood Cliffs, New Jersey, 1995

[2] Dzeroski, S.; Lavrac, N. (eds): Relational Data Mining. Springer, Berlin 2001

Note:

Further information:

No time-table has been prepared for this course

The course is a part of the following study plans:

• Computer Science and Engineering (elective course)