|
|
 |
|
|
|
Matthias Hein, Dr. (Alumnus)
Research Scientist |
|
|
|
|
|
|
|
MPI for Biological Cybernetics
Dept. Schölkopf
Spemannstraße 38
72076 Tübingen
|
| Telephone: | |
+49-681-302 6580 |
| e-mail: | |
hein@cs.uni-sb.de |
|
|
|
|
|
|
|
| |
Since 1.6.2007 I am assistant professor (Juniorprofessor) at the computer science departement at Saarland university. The link
to my new homepage: Link
My new address:
Informatik
Universitaet des Saarlandes
Postfach 151150
D - 66041 Saarbruecken
My main interest lies in the design and the
theoretical analysis of machine learning algorithms, in particular semi-supervised learning and kernel-based algorithms.
Recent research topics:
- Convergence analysis of the continuum limit
of the so called normalized and unnormalized
graph Laplacian:
Special emphasis is laid on the control of the
influence of the density and the case where
the data lies on a submanifold.
- Extension of the Support Vector Machine (SVM)
to arbitrary metric spaces:
Not any metric can be used in SVMs. This is
related to the problem that not all symmetric
functions lead to positive definite kernels.
- Design of Kernels:
The construction of positive definite kernels
on probability measures is studied with special
emphasis on their properties. This type of
kernels is interesting in many applications.
- Intrinsic dimensionality estimation:
For real-world datasets it is often the case
that not all features are independent so that
the data lies effectively on or close to
a submanifold. The dimension of this manifold
or equivalently the degrees of freedom of the
dataset is an important qualitative information which can be used to choose parameters of the learning algorithm.
Code and datasets are available here.
code and datasets
|
|
|
|
|
|
|
|
