Systems Modelling and Data Analysis + Engineering of Intelligent Systems

Lectures: Tuesdays, C-2, Room 429, 15:30-17:00

1. 02-26: CSP (MSL)
2. 03-05: CSP (MSL)
3. 03-12: CSP (MSL)
4. 03-19: CSP (MSL)
5. 03-26: KRR introduction (GJN)
6. 04-02: LPP (GJN)
7. 04-09: Planning (MSL)
8. 04-16: Problog (MSL)
9. 05-07: KR methods overview (GJN)
10. 05-14: RBS (GJN)
11. 05-21: LOD (GJN)
12. 05-28: DL (GJN)
13. 06-04: Wrap up and summary exam - zeroeth term

• Hakank Links
• MiniZinc
• Hakank on MiniZinc
• SWI-Prolog
• Bayes Networks
• Problog
• ASP: Answer Set Programming
• Fuzzy Logic
• Introduction to Fuzzy Logic
• Description Logics

• AI Book
• Prolog - page with external links
• CS221: Artificial Intelligence
• CS227: Knowledge Representation and Reasoning
• KRR: PL

The following classes will focus on modelling of discrete optimization and constraint satisfaction problems. Student will learn how to represent correctly different problems using constraint programming techniques.

• Lab 1. Constraint Programming: 101
• Lab 2. Constraint Programming: Basic Problems
• Lab 3. Constraint Programming: Basic Techniques
• Lab 4. Constraint Programming: Search Modeling
• Labs 5-6. Constraint Programming: Real Life Problem

The following classes will cover automated planning problems. Student will learn how to represent planning problems using constraint programming and dedicated tools.

1. Automated Planning: 101
2. Automated Planning: PDDL
3. Automated Planning: Fluents

This part of the course will present probabilistic programming — a new programming paradigm meant to model domains uncertainty and imperfect knowledge.

1. Probabilistic Programming: 101
2. Probabilistic Programming: Diagnosis and Prediction
3. Probabilistic Programming: Probabilistic Graphs and Decision Theory

There are three projects to choose from:

1. fox-geese-corn — simple planning problem. It's simple but can prove quite difficult to find the optimal solution.
2. gangs-wars — a basic scheduling problem with a fairly complex objective function. While it can be bothersome to model, the optimal solution should be easy to find with a good solver.
3. job-fair — a scheduling+assignment problem involving preferences. It's the most difficult problem out of these three in terms of optimization, as for the modeling phase it is the most structured one.

All the projects are available via gitlab. Instructions, how to do the projects are included in the README.md files.

The deadline is simply the last class in the semester. While grading I will check:

1. if the model is correct;
2. if the model allows to quickly find a good solution;
3. if the model is comprehensible;
4. what was your work hygiene (how often did you commit, did you contact in case of a problem, etc.)

The projects should be done in pairs (if there is an odd number of students, a single team is allowed to have three members).

There will be held four competitions, open per project + one optimization challenge. The rules are simple - find the best solution for the most difficult problem instance as quickly as possible. You can try any solver/method/algorithm/wizard you like, just find the optimum. The prizes are:

1. exam and lab final test exemption
2. lab test exemption
3. 20% of the total lab points extra

The optimization challenge will use the production-planning problem and will be open to all students. There is only a single catch: a solution has to be good enough (objective function below 26200) to be considered. If nobody gets below 26200, then nobody wins. This problem is very difficult to optimize so you may have to try and experiment with different tools and techniques.

The deadline for the competitions is the last Friday before the test/exam (so winners could relax during the weekend).