Jörn Anemüller's Homepage
News
- April 2006
- During the summer term I am acting as substitute professor for signal
processing, since
Alfred Mertins moved to Lübeck University.
Teaching related info is
located here.
- November 2006
- Course info for the seminar "Elemente der Sprach- und
Mustererkennung" is
located here.
- July 2006
- Paper on convolution modeling of fMRI using complex ICA published in Neurocomputing. See the Publications section.
- March 2006
- Paper on multidimensional ICA model published at the ICA 2006 conference. See the Publications section.
- January 2006
- DIRAC (Detection and Identification of Rare Audio-Visual Cues) started, an Integrated Project funded by EC.
- September 2005
- Paper on the new OLLO (Oldenburg Logatome) speech corpus published, see the Publications section.
OLLO provides speech with natural variabilities and is used in the Medical Physics section to compare speech recognition by humans and machines.
The corpus can be downloaded from http://sirius.physik.uni-oldenburg.de.
- April 2005
- Seminar announcement for Oldenburg University:
Veranstaltung 5.04.430: Independent Component Analysis
Zeit und Ort: Mi 16 - 18, W02 1-156
BEGINN: 20.4.2005 (!)
Veranstalter: Jörn Anemüller
Inhalt:
Independent component analysis (ICA) is a relatively novel technique
for data analysis that has proven useful in many situations where
traditionally approaches like principal component analysis (PCA, aka
Karhunen-Loeve transform) have been used. In a nutshell, ICA assumes
that measured data is generated by several underlying 'source'
processes which are mutually independent, e.g., because they constitue
physically separated systems. While the sources are independent, the
recorded signals are mutually dependent mixes of the sources, where
the mixing may, e.g., be due to signal propagation in a physical
medium such as air. The goal of ICA is to reconstruct the sources
from knowledge of the recorded signal, only. Knowledge about the
mixing system is not assumed to be know a-priori, therefore ICA
methods are often termed as blind source separation (BSS). The
description of measured data in terms of several independent signals
may be regarded as the 'most informative'description of the data,
which makes ICA a suitable analysis tool for many applications.
One application of ICA is acoustic source separation where individual
speaker signals are separated from mixed signals recorded with several
microphones. Further applications cover such diverse areas as data
mining, feature extraction, description of natural signals, stock
market analysis, etc.
The seminar will start out with introducing the basic ICA assumptions
and models. Next, convolutive ICA methods will be introduced which
are used in solving the acoustic source separation problem. Finally,
more general topics will be covered such as nonlinear ICA and ICA for
feature extraction, and applications such as biomedical data analysis
from EEG and fMRI, astronomical data analysis, and financial data
analysis.
Along the way, students will be introduced to several general
techniques such as numerical optimization, neural networks,
information theory and statistical signal processing.
Voraussetzungen:
Grundstudium, interest in real-world data analysis
(physical and non-physical systems)
Literatur:
Hyvärinen, Karhunen, Oja: Independent Component Analysis, Wiley, New
York, 2001
Online introductions:
http://www.cis.hut.fi/projects/ica/book/intro.pdf
http://www.cs.helsinki.fi/u/ahyvarin/whatisica.shtml
http://www.cis.hut.fi/projects/ica/icademo/
http://www.cis.hut.fi/projects/ica/cocktail/cocktail_en.cgi
- October 2004
- Starting October, I will be back in Germany at the Dept. of Physics at Oldenburg University.
- Septermber 2004
-
Two papers to be presented at the ICA 2004 conference in Granada, Spain, are now online in the Publications section.