|Line 77:||Line 77:|
|Reinforcement learning I: basic principles and learning methods (TD-learning). Prediction problem.
|Reinforcement learning I: basic principles and learning methods (TD-learning). Prediction problem. [http://dai.fmph.uniba.sk/courses/ICI/References/ci-rl.4x.pdf slides]
|R&N (2010), ch.21.1-2.
|R&N (2010), ch.21.1-2.
Revision as of 13:43, 8 November 2020
Introduction to Computational Intelligence 2-IKV-115a
The course objectives are to make the students familiar with basic principles of various computational methods of data processing that can commonly be called computational intelligence (CI). This includes mainly bottom-up approaches to solutions of (hard) problems based on various heuristics (soft computing), rather than exact approaches of traditional artificial intelligence based on logic (hard computing). Examples of CI are nature-inspired methods (artificial neural networks, evolutionary algorithms, fuzzy systems), as well as probabilistic methods and reinforcement learning. After the course the students will be able to conceptually understand the important terms and algorithms of CI, and choose appropriate method(s) for a given task. The theoretical lectures are combined with the seminar where the important concepts will be discussed and practical examples will be shown.
|Lecture||Monday||9:00 - 10:30||I-9||Igor Farkaš|
|Seminar||Thursday||14:00 - 15:30||I-9||Endre Hamerlik & Igor Farkaš|
|22.09. 16:30 (special date/time)||What is computational intelligence, basic concepts, relation to artificial intelligence. slides||Craenen & Eiben (2003); wikipedia; R&N (2010), chap.1; Sloman (2002)|
|28.09.||Taxonomy of artificial agents, nature of environments. slides||R&N (2010), chap.2|
|05.10.||Inductive learning via observations, decision trees. Model selection. slides||R&N (2010), ch.18.1-3,18.6; Marsland (2015), ch.12, DT visualization, DT in python, information entropy, bias-variance tradeoff|
|12.10.||Supervised learning in feedforward neural networks (perceptrons), pattern classification, function approximation. slides||R&N (2010), ch.18.2; Marsland (2015), ch.3-4, Engelbrecht (2007), ch.2-3|
|19.10.||Unsupervised (self-organizing) neural networks: feature extraction, data visualization. slides||Marsland (2015), ch.14, Engelbrecht (2007), ch.4|
|26.10.||Statistical learning, probabilistic models. slides||R&N (2010), ch.13,20.1-2|
|02.11.||Q&A - preparation for midterm||Thursday: mid-term test|
|09.11.||Reinforcement learning I: basic principles and learning methods (TD-learning). Prediction problem. slides||R&N (2010), ch.21.1-2.|
|16.11.||Reinforcement learning II (Q, SARSA), actor-critic, control problem, RL for continuous domains.||R&N (2010), ch.21.3-5; Woergoetter & Porr (2008).|
|23.11.||Evolutionary computation: basic concepts, genetic algorithms.||Engelbrecht (2007), ch.8|
|30.11.||Fuzzy systems, fuzzy logic and reasoning.||Engelbrecht (2007), ch.20-21; Zadeh (2007)|
|07.12.||Summary, recap of main concepts, synergies.||Q&A|
Note: Dates refer to lectures, seminars will be on day+3 each week.
- Craenen B., Eiben A. (2003): Computational Intelligence. In: Encyclopedia of Life Support Sciences, EOLSS Publishers Co.
- Engelbrecht A. (2007). Computational Intelligence: An Introduction (2nd ed.), John Willey & Sons.
- Russell S., Norwig P. (2010). Artificial Intelligence: A Modern Approach, (3rd ed.), Prentice Hall. Available in the faculty library.
- Marsland S. (2015). Machine Learning: An Algorithmic Perspective, (2nd ed.), CRC Press.
- Sloman A. (2002). The Irrelevance of Turing Machines to AI. In Scheutz M. (ed.): Computationalism: New Directions, MIT Press, Cambridge, MA, pp. 87–127.
- Woergoetter F., Porr B. (2008). Reinforcement learning, Scholarpedia, 3(3):1448.
- Zadeh L. (2007). Fuzzy logic, Scholarpedia, 3(3):1766.
- Active participation during the lectures/exercises (35%): 15 for lectures, 20 for exercises.
- Written mid-term test (30%).
- Final written-oral exam (30%): you will choose 3 questions, minimum of 1/3 of all points required.
- Small final project (10%) = implementation of a small neural network (using an existing Python library) and writing a short report. Note: even without this, the student can still get maximum points if s/he has performed very actively.
- Overall grading: A (>90%), B (>80%), C (>70%), D (>60%), E (>50%), Fx (otherwise).