Skip to content. | Skip to navigation

Sections
Personal tools
You are here: Home Staff Sascha Martin Welten, M.Sc.

Sascha Martin Welten, M.Sc. - Theses


Master
Blockchain-based Swarm Learning Framework in Personal Health Train - Running
Advanced analytics could power the data collected from numerous sources, both from healthcare institutions, or generated by individuals themselves via apps and devices, and lead to innovations in treatment and diagnosis of diseases; improve the care given to the patient, and empower citizens to participate in the decision-making process regarding their own health and well-being. With the world-widely emergence of data protection legislation, e.g., General Data Protection Regulation (GDPR) in the European Union, society is more aware of privacy. The sensitive nature of health data prohibits healthcare organizations from collecting and sharing the data. Traditional data analytics methods based on data centralization become less feasible. Distributed approaches shifting algorithms instead of data are solutions to comply with privacy protection regulations.
Bachelor, Master
Performing Distributed Analytics on Decentralised Geo- and Hydrological Data - Finished
Completed by in 2021; Supervised by Prof. Dr. Stefan Decker, Prof. Dr. Holger Schüttrumpf; Advisor(s): Sascha Martin Welten, M.Sc., Lennart Schelter, Julian Hofmann
In times of increasing weather extremes, weather forecasts and early warning systems for extreme weathers have gained great promise to prevent property damage or to mitigate the risk to the civilian population. Current approaches are based on different data types, such as geo basic data, weather forecasts or other hydrological data, in several data sources (Salas et al., 2020; Kreklow et al., 2020; Hofmann and Schüttrumpf, 2019) However, these approaches rely on data centralisation, which can pose several challenges. First, the centralisation process itself could act as a bottleneck since vast amounts of spatiotemporal data from multiple data sources are transmitted to one single location. Second, the data has to be harmonised locally such that lightweight data analysis tasks or even data model training can be conducted. Lastly, depending on the data type, well-known data privacy regulations hinder institutions to share their data. One solution to circumvent these immanent problems of data centralisation could be Distributed Analytics (DA) approaches. DA poses a paradigm shift to address the mentioned challenges by bringing the analysis algorithm to the data instead of vice versa. By design, data stays within institutional borders and the data owner keeps the sovereignty over the data.
Bachelor
Development of Generative Models Trained on Decentralised Geo- and Hydrological Data - Finished
Completed by in 2021; Supervised by Prof. Dr. Stefan Decker, Prof. Dr. Holger Schüttrumpf; Advisor(s): Sascha Martin Welten, M.Sc., Julian Hofmann, Lennart Schelter
In times of increasing weather extremes, weather forecasts and early warning systems for extreme weathers have gained great promise to prevent property damage or to mitigate the risk to the civilian population. Current approaches are based on different data types, such as geo basic data, weather forecasts or other hydrological data, in several data sources (Salas et al., 2020; Kreklow et al., 2020; Hofmann and Schüttrumpf, 2019) However, these approaches rely on data centralisation, which can pose several challenges. First, the centralisation process itself could act as a bottleneck since vast amounts of spatiotemporal data from multiple data sources are transmitted to one single location. Second, depending on the data type, well-known data privacy regulations hinder institutions to share their data. Lastly, data sets suffer from sparsity and incomplete features, which makes data analysis difficult. In particular, weather datasets provide high spatiotemporal data whereas hydrological well data or geobase data are generally incomplete.
Master
Model Training through Curiosity-based Latent Space Exploration on Decentralized Data - Finished
Completed by in 2021; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Sascha Martin Welten, M.Sc.
Recent developments in privacy rights have highlighted the need for methods capable of training deep learning models on non IID data. In this thesis, we present an approach to train deep learning models on distributed datasets without gathering the data in a central location. We introduce an experimental setup to simulate such distributed datasets with a non-distributed one and show our ndings on MNIST, Cifar-10 and the ISIC challenge. Current methods based on generative models su er from mode collapse when trained on non IID data. As such we develop an approach based on concepts from Distributed Analytics, Active Learning and Reinforcement Learning. Using the idea of measuring informativeness to select samples from Active Learning and the Curiosity Module from Reinforcement Learning to measure said informativeness, we show that curiosity can be used to improve sample eciency and prediction performance. We also show that models trained without our method fail to produce any usable results on data which is both non IID and has a high class imbalance.
Master
A Smart Routing Algorithm for the Personal Health Train (PHT) - Finished
Completed by in ; Supervised by Prof. Dr. Stefan Decker; Advisor(s): Sascha Martin Welten, M.Sc., Dr. Oya Deniz Beyan
In the context of this master thesis, the student should investigate on different routing heuristics for the PHT in a distributed Machine Learning/Deep Learning setting.