• Home
• What is IK?
• Focus Theme 2011
• Organization
• Program
  • IK 2011 Schedule
• Poster and Demo Session
• Location
• Stipends
• Registration
• Contact

Speaker
Marc Toussaint

Content

There are at least two reasons for why a cognitive scientist,
neuroscientist and roboticist needs to know about Machine Learning:
First, Machine Learning has developed a large set of tools for
statistical data analysis which have become indispensable for
modelling empirical data in all disciplines. For instance, Machine
Learning methods allow a neuroscientist to clearly quantify whether
there exists mutual information (a form of correlation) between
certain brain activity patterns and a certain behavioral feature. This
is the view of Machine Learning as an analysis tool.

Second, learning and decision making are among the most interesting
aspects of intelligent behavior in humans and animals. Machine
Learning provides a computational view on what learning in principle
means (e.g., in an information theoretic or Bayesian sense). It
clarifies the crucial role of priors and/or regularization for any
type of learning. Thereby it allows us to formulate computational
models of human and animal learning, decision making and goal-directed
behavior itself. To give some examples in current research: decision
making has been modeled as Bayesian inference; bounded rationality has
been modelled as Markov Chain Monte Carlo process; response times in
decision making have been modelled as sequential ratio probability
tests; goal-directed behavior and motor control has been modelled as
Bayesian inference, etc. The trend here is that concepts from
cognitive science and psychology are formalized with computational
models in the framework of Machine Learning. This is the view of
Machine Learning as a computational framework to describe learning and
decision making itself.

In this course I will introduce to the basics of Machine Learning,
focussing on those aspects that seem particularly relevant from the
''behavioral'' point of view, i.e., for cognitive science,
neuroscience and robotics. I will roughly cover the following topics,
not necessarily in this order:

• Classical Machine Learning: PCA, SVMs, etc.
• The Bayesian view: The role of priors, Bayesian inference and
  Bayesian networks, fully Bayesian vs. likelihood maximization,
  expectation maximization & Free Energy, etc.
• Decision making as inference, Bayesian methods in Cognitive
  Sciences and Perception, neural substrates to perform Bayesian inference?
• Reinforcement Learning, goal-directed vs. habitual behavior, applications in Robotics.

Disciplines
Machine Learning, Bayes, Applications in Cognitive Science, Robotics,
and Neuroscience.

CV
Marc Toussaint is heading the Machine Learning and Robotics Lab at FU
Berlin, Germany. See http://user.cs.tu-berlin.de/~mtoussai/ for more
details.