The content of this course is independent of the content of the
course machine learning. Both courses count as core courses for
the Computational Intelligence Catalogue, and are part of the
course catalogue for Doctoral Students.
This course presents the most successful methods for designing
systems that learn autonomously, i.e., without a supervisor who
tells the system at every trial what the ''right'' answer or
action would have been. It is not surprising that many currently
existing methods for autonomous learning are inspired by
learning processes in biological organism, since most of their
learning has to take place without a supervisor.
Also the most successful learning algorithms in real-world
robots are reinforcement learning algorithms,
The course will present the best currently existing mathematical
models and algorithmic solutions for autonomous learning,
including recent work from Google Deep mind on learning to win
in Atari games, and in the game Go.
We will start with Genetic Algorithms that mimick learning on
the time-scale of evolution, and also discuss Simulated
Annealing as an alternative. Then we will present learning
strategies for learning to act in an unknown environment, with
the goal to maximize rewards (Reinforcement Learning). For the
more recent results from Google Deep mind we will discuss the
combination of Deep Learning with Reinforcement Learning.