Tobias Andreas Fritz M.Sc.

Research Area:

Tobias' current research area is in the domain of temporal graph neural networks. Here, he is especially interested in applying these for the task of cryptocurrency blockchain transaction graph analysis for ransomware payment tracking as well as anomaly detection in network security problems. Another focus lies on the optimization of performance, runtime and memory efficiency of temporal graph neural networks. Further, he is interested in AI agents based on large language models and their application to autonomous network management.

Before starting his doctorate, he worked for one year as a software developer. He received his master's degree in mathematics from the Technical University of Munich and his bachelor's degree in mathematics with a minor in computer science from the Ludwig-Maximilians-Universität. During his studies he worked in research departments at Infineon Technologies AG and Allianz Global Investors GmbH for about 3 years.

Publications:

Tobias Fritz, Alexander Schwankner, Jan-Hendrik Wissing, Robin Buchta, Gabi Dreo Rodosek. May 2025. Granomaly: A Framework for Anomaly Detection in 5G Core Network Control Plane Traffic with Temporal Graph Neural Networks. Accepted for Publication @ Network Operations and Management Symposium (NOMS) 2025.

Robin Buchta, Tobias Fritz, Carsten Kleiner, Felix Heine, Gabi Dreo Rodosek. May 2024. One-Class Learning on Temporal Graphs for Attack Detection in Cyber-Physical Systems, Network Operations and Management Symposium, Workshop on Analytics for Network and Service Management (NOMS AnNet) 2024.

Tobias Fritz. September 2023. Leveraging tree-structured Graphs in Graph Neural Networks for Fake News Detection, Poster @ European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML  PKDD) 2023

L. Servadei, E. Zennaro, T. Fritz, K. Devarajegowda, W. Ecker, R. Wille. November 2019.  Using Machine Learning for Predicting Area and Firmware Metrics of Hardware Designs from Abstract Specifications. In Microprocessors and Microsystems,  https://doi.org/10.1016/j.micpro.2019.102853

Seminar/Bachelor/Master Thesis Topics:

Bachelor / Master Thesis:

Efficient Implementation of Machine Learning based Anomaly Detection in 5G Networks within NWDAF: The 3rd Generation Partnership Project introduced a service-based architecture in the 5G core network, enabling flexible communication through modular network functions and the separation of control and user planes. While the integration of machine learning based anomaly detection methods via the Network Data Analytics Function (NWDAF) has improved security, challenges persist with runtime efficiency of these methods. Therefore, in this thesis, you will analyze how performance improves when moving the anomaly detection from package-level to more abstract network features.

Comparison of different Machine Learning Methods for Anomaly Detection in 5G Network traffic: The 3rd Generation Partnership Project introduced a service-based architecture in the 5G core network, enabling flexible communication through modular network functions and the separation of control and user planes. In this thesis, you will use our 5G simulation framework to compare different Machine Learning methods like Temporal Graph Neural Networks, Convolutional Neural networks, etc. with respect to their performance to detect attacks in the network traffic.

Detection of Ethereum Scam Tokens via Temporal Graph Learning: Blockchain networks require every transaction to be verified and recorded publicly. For example, the entire history of Bitcoin transactions—amounting to roughly 500GB of data—includes publicly available sender and receiver addresses, transaction amount and timestamp. This transparency enables modeling transactions as temporal graphs. Recently, Temporal Graph Neural Networks (TGNNs) have been developed for performing machine learning on such complex data structures. This thesis focuses on detecting Ethereum scam tokens (including rug pulls, honeypot schemes, and Ponzi tokens) using TGNNs. With crypto scams causing hundreds of millions in damages each year, the public availability of transaction data offers a promising avenue for forensic investigation and fraud detection. You will apply different TGNN methods to an existing Ethereum dataset. The objectives are to train a model to accurately identify scam tokens on this dataset and subsequently evaluate its performance on more recent Ethereum transactions.

Automated Analysis of News Websites – Integration of Web Crawlers and Large Language Models: This thesis offers the opportunity to develop and implement an innovative framework for the automated analysis of news websites. The project is divided into two main components: 

1. Web Crawler Development:

• Design, implementation, and optimization of a web crawler to systematically extract relevant content from selected news websites.
• Evaluation of data collection techniques with respect to efficiency, scalability, and quality.

2. Data Structuring and LLM Integration:

• Organization and preprocessing of the extracted data to ensure it is optimally formatted for further analysis.
• Application and assessment of state-of-the-art Large Language Models (LLMs) for automated information retrieval and processing.

The goal of this project is to advance automated information systems in the media sector by leveraging interdisciplinary expertise in web data mining, natural language processing, and machine learning.

Seminar Thesis: