Reducing the Effort for Data Annotation: Contributions to Weakly Supervised Deep Learning

MCML Authors

Jann Goschenhofer

Dr.

* Former Member

→ Group Bernd Bischl
 Statistical Learning and Data Science

Abstract

This thesis addresses methods for training machine learning models with limited labeled data, focusing on semi-supervised, positive unlabeled, constrained clustering, and transfer learning. It explores deep semi-supervised learning, particularly in time series and medical imaging contexts, and investigates positive unlabeled learning methods that utilize predictive uncertainty for self-training. The thesis also introduces weakly supervised learning for constrained clustering, combining it with semi-supervised approaches, and applies transfer learning to tasks with varying granularity in medical domains. (Shortened).

phdthesis

---

Dissertation

LMU München. Dec. 2023

---

Authors

J. Goschenhofer

---

Links

DOI

---

Research Area

 A1 | Statistical Foundations & Explainability