Symbolic Machine Learning

Semestr: Summer

Range: 2+2c

Completion:

Credits: 6

Programme type:

Study form: Fulltime

Course language: English

Summary:

The course will explain methods through which an intelligent agent can learn, that is, improve its behavior from observed data and background knowledge. The learning scenarios will include on-line learning and learning from i.i.d. data (along with the PAC theory of learnability), as well as the active and reinforcement learning scenarios. Agent's knowledge will be represented through the language of logic and through graphical models. The course is given in English to all students.

Keywords:

Course syllabus:

1. Agent-environment model, interaction principles
2. Concept learning, on-line learnability, version space
3. Learning from i.i.d. data, PAC-learnability, VC-dimension
4. Learnability of propositional-logic concepts
5. Learning a graphical probabilistic model
6. Learning a graphical probabilistic model (2)
7. PAC-learning predicate-logic CNF
8. Learning predicate-logic clauses
9. Learning a relational graphical probabilistic model
10. Learning a relational graphical probabilistic model (2)
11. Active learning
12. Reinforcemenent learning
13. Reinforcemenent Learning (2)
14. Solomonoff induction and universal AI

Seminar syllabus:

1 Introduction, Python environment, entrance test
2 PEAS agent model, environment properties, input/output types of machine learning models, demos
3 Learning conjunctive and disjunctive concepts
4 Assignment of the first student project
5 Bayesian Networks - semantics
6 Bayesian Networks - inference
7 Assignment of the second student project
8 Inductive Logic Programming - learning from interpretations
9 Inductive Logic Programming - learning from clauses
10 ILP, Q&A
11 Reinforcement learning, demos and introduction
12 Assignment of the fourth project
13 Passive reinforcement learning agents, TD methods
14 Reserve, assessment

Literature:

Course textbook available at https://cw.fel.cvut.cz/wiki/courses/b4m36smu/lectures

Stuart Russell and Peter Norvig: Artificial Intelligence: A Modern Approach, Prentice Hall 2010

Luc De Raedt: Logical and Relational Learning, Springer 2008

Marcus Hutter: Universal artificial intelligence, Springer 2005

Examiners:

Lecturers:

Instructors: