Stefan Uhlich

I am a Principal Engineer at Sony Stuttgart Technology Center where I work in the area of signal processing and deep learning.
Before working at Sony I was a research assistant at the Institute of Signal Processing and System Theory where I obtained my PhD in statistical signal processing.

E-mail: uhlich . s ( at ) gmail . com

Publications

Articles

Thesis

Lectures

  • Detection and Pattern Recognition [Lecture contents]
    Summer Term 2011, University of Stuttgart

  • Matrix Computations in Signal Processing and Machine Learning [Lecture contents]
    since Winter Term 2015/2016, University of Stuttgart

Software (Matlab Source Code)

  • Estimator Family Toolbox for Nonstandard Loss Functions
    This Matlab toolbox contains an implementation of the parametric estimator family which was proposed here.
    Estimator Family Toolbox V1.0 (May 2011)

  • Multiple Description Toolbox
    This Matlab toolbox contains several scripts to determine the optimal correlating transform. Multiple description coding is used to safely transmit data packets (descriptions) over erasure channels. This toolbox contains a demo together with scripts to calculate the correlating transform. It builds upon the work from Romano and was extended to handle also the case, that redundant descriptions were transmitted.
    MDC Toolbox V1.2 (November 2007)
    Keywords: Multiple description coding (MDC), Correlating transform

  • Classification Toolbox
    This Matlab toolbox contains several classifiers for pattern recognition together with a documentation. Additionally, it includes functions for feature selection and feature extraction.
    Classification Toolbox V1.5 (September 2010)
    Keywords: Classification, Neural network, Linear discriminant function, Bayes classifier, Fisher LDA, PCA, SFFS

  • Estimation in a linear Gaussian Model with Ellipsoidal Constraints
    This is a Matlab implementation of various estimators for the estimation of an unknown parameter vector in a linear Gaussian model with ellipsoidal constraints. The signal model is x = H * theta + z where theta is assumed to lie in or on an ellipsis and it also contains the MMSE estimator that we proposed here.
    Version 1.0 (September 2008)
    Keywords: Parameter estimation, Minimum mean squared error estimation, Restricted parameter space