Representation Learning for Domain Adaptation and Cross-Modal Retrieval: In the Context of Online Handwriting Recognition and Visual Self-Localization

MCML Authors

Felix Ott

Dr.

* Former Member

→ Group Bernd Bischl
 Statistical Learning and Data Science

Abstract

This thesis focuses on domain adaptation and cross-modal retrieval to address the challenges posed by domain shifts in machine learning applications. Specifically, it explores techniques for online handwriting recognition and visual self-localization. For handwriting recognition, the study uses deep metric learning and optimal transport to reduce domain shifts between different writing styles and writing modalities, while for visual self-localization, it enhances pose prediction through auxiliary tasks and representation learning fusion techniques to improve accuracy across sensor modalities. (Shortened.)

phdthesis

---

Dissertation

LMU München. Jul. 2023

---

Authors

F. Ott

---

Links

DOI

---

Research Area

 A1 | Statistical Foundations & Explainability