You are here

Warning message

Attention! This event has already passed.

Identification and use of nonparametric noise models extracted from overlapping subrecords

Friday, 25 September, 2009 - 15:00
Campus: Brussels Humanities, Sciences & Engineering campus
Kurt Barbe
phd defence

Models are very important for studying, understanding, controlling, predicting and optimizing
observed phenomena. Since, these models should grasp the observed reality as good as possible,
they should be based on real life data. Unfortunately life is not ideal and measurements are prone to
errors. These errors can completely blur our view of the observed phenomena. To optimaly reduce
the influence of these measurement errors on the model one needs to quantify the behavior and
nature of these errors, this is done with nonparametric noise models.

To develop these noise models, we often need long and costly experiments. The classical theory
requires at least six repetitions of the same experiments to ensure the good quality of the noise
models. Besides that, the subexperiments should be long enough to obtain a good match between
model and observations. The main contribution of this thesis is to obtain good quality noise models
with only 2 repetitions. Recycling is very important nowadays, even in the modeling world! So the
strategy used in this thesis allows one to recycle measurements points from both repetitions. This
results in overlapping subrecords. This recycling strategy works remarkably well and its performance
is studied in full detail.