Vær opmærksom på at dette website indeholder et arkiv med historiske data. Det aktuelle kursuskatalog findes på kursuskatalog.au.dk

AU kursuskatalog arkiv

[Forside] [Hovedområder] [Perioder] [Udannelser] [Alle kurser på en side]

Ikke-lineær signalbehandling og mønstergenkendelse (TINONS1U01) (Q4) ( forår 2011 - 5 ECTS )

Rammer for udbud

  • Uddannelsessprog: engelsk
  • Niveau: Graduate course
  • Semester/kvarter: Q4
  • Timer per uge: 4
  • Deltagerbegrænsning: None
  • Undervisningssted: Århus
  • Hovedområde: Ingeniørhøjskolen
  • Udbud ID: 29765

Formål

The aim of the course is to give participants insight into methods for Nonlinear Signal Processing and Pattern Recognition of real world signals such as music, images/video or physiological signal from the human body. Participants will gain experience with nonlinear models of such signals, where techniques from estimation­/detection theory and machine learning will be used for identification of suitable models and further analysis and decision making based on the signals. Participants will obtain practical experience, through work on a smaller number of topic reports.

Indhold

The course will present nonlinear methods for analyzing and decision making based on real world signals. Techniques from estimation/detection theory and machine learning will be presented and applied on specific cases (tentative 4 mini projects).

  1. Gesture Recognition (accelerometer signals).
  2. Music Genre Classification (sound signals).
  3. Face Recognition (image/video signal).

Human Computer Interface based on brain signals (physiological signal from the human body).

Faglige forudsætninger

Students most know of signal analyses and signal processing at the level of a bachelor degree in electronic engineering or software engineering supplemented with the course in Advanced Signal Processing and Analysis. Bachelors in computer science, physics or mathematics, with primary knowledge of stochastic signal analyses and signal processing will also be able to attend.

Underviser

Peter Ahrendt and Henrik Karstoft

Undervisnings- og arbejdsform

Lectures, labs and mini reports.

Litteratur

To be announced

Eksamensterminer

June, re-examination after appointment with lecturer

Udbyder

Aarhus School of Engineering (ASE)

Tilmelding til undervisning

http://mit.au.dk/

Læringsmål

The participants must at the end of the course be able to:

  • Explain basic terminology such as supervised/unsupervised learning, likelihood, the bias-variance relation and discriminative/generative models.
  • Compare , relate and analyze different methods for feature extraction and feature selection on real world signals.
  • Relate and compare Nonlinear Signal Processing to previously learned material such as linear FIR/IIR digital filters and adaptive filter theory.
  • Design and evaluate algorithms for Linear Regression and Classification on real world signals.
  • Apply and explain Artificial Neural Networks on real world signals.
  • Apply and explain Gaussian Mixture Models and EM-algorithm on real world signals.
  • Apply and explain Sampling Methods on real world signals.
  • Apply and explain Principal Component Analyses on real world signals.
  • Apply and explain Hidden Markov Models on real world signals.

Bedømmelse

Oral examination (30. min) based on curriculum questions (67%) and mini report (33%).  Grades on the 7-scale. External examiner.