Profile picture
Understanding Cochlear Implant Outcome Variability using Big Data and Machine Learning

Learn more about the project here.

Supervised by Prof. Dr. tech. Wolfgang Nejdl, L3S Research Center Hannover.

Master's Thesis
Application of Graph Structure Learning in Biological Data Analysis

Graph Neural Networks (GNNs) performance relies heavily on the input graph structure quality. Noisy and incomplete graphs often lead to poor model output. Graph Structure Learning (GSL) is a procedure to improve the graph structure by removing and adding edges and/or changing edge weights of the original graph. Since the biological data sets are often incomplete and contain many false positives, GSL can be used to improve the quality of the original graph which in turn can lead to an improvement of the models predictive performance. In this thesis, I examine the effectiveness of GSL on the miRNA-disease association prediction problem. MicroRNAs (miRNAs) are a type of non-coding RNAs that can regulate protein coding genes (PCGs) expression. Disruption in miRNA expression can result in disruption of protein coding gene expression which can lead to disease conditions. Taken into account the complex inter-relationship between miRNA, disease and protein coding genes, I formulate this task as a classification problem in which I employ graph representation learning techniques on a heterogeneous network constructed from the interrelations between miRNAs, PCGs and diseases to learn the input representation. I propose a Network Optimization framework for MicroRNA-Disease Association prediction (NOMiDiA). NOMiDiA relies on GSL and learns an optimized graph by incorporating information from the original graph structure and the derived feature similarity of each node pairs. This optimized graph is fed to a GNN to infer associations between miRNAs and diseases. The proposed approach acquires better performance on benchmarked data sets compared to the existing baseline.

Supervised by Dr. Megha Khosla and Ngan Thi Dong, L3S Research Center Hannover.

Bachelor's Thesis
Usable Network Security Configuration for Android Developers

Research has found issues that developers are facing while implementing secure applications in Android Studio, even after the Network Security Configuration was introduced and updated to more secure default configurations some flaws weren’t addressed in the latest version1. In my bachelor thesis I addressed these problems and proposed a solution in form of a plug-in for Android Studio. The features of the plug-in cover wrong root tags, allowed cleartext traffic, malformed domains and self-signed certific- ates by providing code highlighting, mouse over texts and quick fixes for the developer. Therefore this plug-in can help developers to easily detect security issues and assists in correcting them. To meet my design goals of providing help that is easy to understand and visible but not distracting, I conducted a focus group while developing the plug-in.

Supervised by Prof. Sascha Fahl and Nicolas Huaman, TeamUSEC at Leibniz University Hannover.