Stalph P. Analysis and Design of Machine Learning Techniques: Evolutionary Solutions for Regression, Prediction, and Control Problems

Springer, 2014. — 162 p. — ISBN: 978-3-658-04936-2, e-ISBN: 978-3-658-04937-9.Manipulating or grasping objects seems like a trivial task for humans, as these are motor skills of everyday life. However, motor skills are not easy to learn – babies require several month to develop proper grasping skills. Learning motor skills is also an active research topic in robotics. However, most solutions are optimized for industrial applications and, thus, few are plausible explanations for learning in the human brain.
The fundamental challenge, that motivates this research work, originates from the cognitive science: How do humans learn their motor skills? This work makes a connection between robotics and cognitive sciences by analyzing motor skill learning in well defined, analytically tractable scenarios using algorithmic implementations that could be found in human brain structures – at least to some extent. However, the work is on the technical side of the two research fields and oriented towards robotics.Introduction and Motivation
I Background
Introduction to Function Approximation and Regression
Elementary Features of Local Learning Algorithms
Algorithmic Description of XCSF
II Analysis and Enhancements of XCSF
How and Why XCSF works
Evolutionary Challenges for XCSF
III Control Applications in Robotics
Basics of Kinematic Robot Control
Learning Directional Control of an Anthropomorphic Arm
Visual Servoing for the iCub
Summary and Conclusion
A A one-dimensional Toy Problem for Regression