KIT Career ServiceStudierendeAbschlussarbeiten

Implicit Diffeomorphisms Using Gaussian Mixtures

Forschungsthema/Bereich
Math, Computer Science
Typ der Abschlussarbeit
Bachelor / Master
Startzeitpunkt
-
Bewerbungsschluss
31.03.2028
Dauer der Arbeit
4 moths for a BSc and 6 months for an MSc

Beschreibung

Sometimes you need a probabilistic representation of a differentiable mapping, and for many reasons having it in Gaussian Mixture form is useful. Gaussian Mixtures are commonly fitted to data using expectation maximization. The algorithm is not guaranteed to converge to the optimum, is data hungry, and can be computationally expensive. The project will explore the construction, implementation, and evaluation of such geometry-induced Gaussian mixtures.

Possible directions include:
• Constructing local Gaussian components from a partition of the parameter domain.
• Propagating local probability density function through the push forward of the mapping.
• Investigating the resulting joint, forward, and inverse probabilistic representations.
• Studying the influence of the partition and local approximation on the quality of the representation.

Voraussetzung

Voraussetzungen an Studierende
  • • Firm grasp over Linear/Multilinear Algebra and Analysis
  • • Good understanding of Probability and Statistics
  • • PyTorch or an equivalent NumPy-like framework
  • • Some exposure to Differential Geometry is advantageous
  • • Interest in numerical methods and mathematical modelling

Studiengangsbereiche
  • Ingenieurwissenschaften
    Elektrotechnik & Informationstechnik
    Informatik
    Computer Science
    Electrical Engineering and Information Technology
  • Naturwissenschaften und Technik
    Mathematik
    Physik
    Technomathematik
    Computational and Data Sience
    Physics
    Wirtschaftsmathematik
  • Sonstiges z.B. Lehramt
    Informatik
    Mathematik
    Physik
  • Wirtschafts- und Rechtswissenschaften
    Wirtschaftsinformatik


Betreuung

Titel, Vorname, Name
Ali Darijani
Organisationseinheit
Fraunhofer IOSB/KIT CS
E-Mail Adresse
ali.darijani@iosb.fraunhofer.de
Link zur eigenen Homepage/Personenseite
Website

Bewerbung per E-Mail

Bewerbungsunterlagen
  • Lebenslauf
  • Notenauszug
  • Whatever that you like me to see!

E-Mail Adresse für die Bewerbung
Senden Sie die oben genannten Bewerbungsunterlagen bitte per Mail an ali.darijani@iosb.fraunhofer.de


Zurück