Connectionists: PhD Studentship 'Random Tree Kernels'

[Raphael.Maree.]

Dear colleagues,

Please find below a PhD research studentship position 
at the University of Liège, Belgium.


----------------------------------------------------------------------------
‘Kernel properties of Random Tree Models’

Random tree models have been extensively developed in the field of
Machine Learning in the recent years. Instances of such methods are for
example bagging, random subspace, random forests, and extra-trees.

Random trees can be generated efficiently and the combination of large
sets of random trees generally leads to accurate models. Under some
hypothesis on the random distribution of trees, it is possible to
characterize analytically the approximation produced by an infinite
ensemble of trees. These models can also be interpreted as Kernel
interpolators where the kernel is a mixture of randomized piece-wise
constant kernels. Different random distribution of the tree models lead
to different geometrical and statistical properties of the resulting
models and kernels.

The subject of this thesis project concerns the theoretical study of the
properties of various Random Tree Models in order to improve our basic
understanding of these methods and create new algorithms with
pre-specified properties.

Subject areas: machine learning, random processes, statistics, geometry
and functional analysis

References:
1. Zhao, G., `A new perspective on classification'. Ph.D. thesis, Utah
State University, Department of Mathematics and Statistics, 2000.

2. Breiman, L., `Some infinity theory for predictor ensembles'.
Technical Report 579, University of California, Department of Statistics,
2000.

3. Lin, Y. and Y. Jeon, `Random forests and adaptive nearest neighbors'.
Technical Report 1055, University of Wisconsin, Department of Statistics,
2002.

4. Breiman, L., `Consistency for a simple model of random forests',
Technical Report 579, University of California, Department of Statistics,
2004

5. P. Geurts, ‘Contributions to decision tree induction: bias/variance
tradeoff and time series classification’, PhD Thesis, University of
Liège, Department of Electrical Engineering and Computer Science’, 2002.


Contact: Prof. Louis Wehenkel   L.Wehenkel at ulg.ac.be
http://www.montefiore.ulg.ac.be/services/stochastic/
----------------------------------------------------------------------------