Vrije Universiteit Amsterdam
Amsterdam, The Netherlands
Area of Study
Taught In English
Course Level Recommendations
ISA offers course level recommendations in an effort to facilitate the determination of course levels by credential evaluators.We advice each institution to have their own credentials evaluator make the final decision regrading course levels.
Recommended U.S. Semester Credits3
Recommended U.S. Quarter Units4
Hours & Credits
• Knowledge and understanding of:
◦ Complex optimisation problems (graphical discrete and continuous optimisation problems
◦ Exact optimisation algorithms
◦ Heuristic optimisation algorithms
• Applying knowledge and understanding
◦ Applying heuristic optmisation algorithms to combinatorial (discrete) optimisation problems
◦ Applying optimisation algorithms to learn a controller for a robot
• Making judgements
◦ About which algorithms to apply for complex optimisation problems
◦ Writing a report on comparing algorithms in the style of an experiments section of a scientific paper.
In the course Computational Intelligence, we will focus on a part of artificial intelligence beyond simple supervised (machine) learning. Specifically, we will attempt to solve problems that are extremely hard to solve, and therefore require algorithmic measures to have a shot at finding a good solution. In the first part, we will focus on graph-based (combinatorial optimisation) problems with discrete input domains. These problems pop up everywhere in computer science and AI, from task scheduling on chips, to coordination in robot soccer, to parking and scheduling maintenance tasks for trains at the NS. We will first show how exact algoritms for such problems (such as variable elimination and AND/OR tree search) can find optimal solutions for small program instances but do not scale up to larger problems in terms of either memory or computational requires, or both. Therefore, we will need to open our approximate-algorithm toolbox. We will introduce different classes of algorithms that can be used to tackle these problems: specifically, (multi-start, iterative, genetic) local search, evolutionary algorithms, and swarm optimisation. Then, in the second part, we will move to contnous optimsation problems, such as finding a good controller for a morphologically evolved robot. Such continuous input. For example, similar ideas underpin stochastic gradient decent and local search algorithms, and CMA-ES is an evolutionary startegy that can be used effectively in continuous domains. We will further look at the use of neural networks, and Bayesian optimisation for such problems.
Lectures and practical assignments
TYPE OF ASSESSMENT
Final exam and practical assignments
Courses and course hours of instruction are subject to change.
Some courses may require additional fees.