Katharina Dietz M. Sc.

Telefon	(0931) 31-87343
E-Mail	katharina.dietz@informatik.uni-wuerzburg.de
Raum	A212
Anschrift	Lehrstuhl für Informatik III Am Hubland D-97074 Würzburg

Contents: Overview | Working Group | Projects | Teaching | Publications

Lectures (Teaching Assistant)

Leistungsbewertung verteilter Systeme (LBVS) (SS24)
Leistungsbewertung verteilter Systeme (LBVS) (SS23)
Leistungsbewertung verteilter Systeme (LBVS) (SS22)
Steuerungsprinzipien moderner Kommunikationssysteme (SKS) (WS20/21)

Seminar Talks (Tutor)

User-centric Communication Networks (UCN) (SS24)

D. L.: Explainable AI (XAI) for Anomaly Detection in Communication Networks
E. N.: Personas in Cybersecurity and/or Network Monitoring Contexts

User-centric Communication Networks (UCN) (SS23)

T. S.: Transfer Learning Approaches for Network Security/Intrusion Detection

User-centric Communication Networks (UCN) (WS22/23)

M. E.: MimicNet: Fast Performance Estimates for Data Center Networks with Machine Learning (based on the SIGCOMM Paper by Zhang et al.)

User-centric Communication Networks (UCN) (SS22)

P. A.: Graph Neural Networks for Malware Detection

User-centric Communication Networks (UCN) (WS21/22)

Y. S.: Firewall Fingerprinting
A. N.: Malware Fingerprinting

User-centric Communication Networks (UCN) (SS21)

M. S.: QoE and Privacy
B. H.: QoE and Security

User-centric Communication Networks (UCN) (WS20/21)

P. K.: Reinforcement Learning for Traffic Routing
M. E.: Reinforcement Learning for Network Security

Next Generation Networks (NGN) (SS20)

L. R.: SDN in the Datacenter: Usecases and Challenges
A. R.: SDN outside the Datacenter: SD-WAN and Co.

Supervised Theses

1.
Ziegler, K.: Comparison of Explainable AI (XAI) Techniques for Flow-based Anomaly Detection in Communication Networks, (2024).
- [ BibTeX ]
- [ EndNote ]
@mastersthesis{info3-bachelor-2024-2, author = {Ziegler, Kilian}, keywords = {ucn}, month = {07}, school = {University of Würzburg}, title = {Comparison of Explainable AI (XAI) Techniques for Flow-based Anomaly Detection in Communication Networks}, year = 2024 }

%0 Thesis %1 info3-bachelor-2024-2 %A Ziegler, Kilian %D 2024 %T Comparison of Explainable AI (XAI) Techniques for Flow-based Anomaly Detection in Communication Networks
1.
Gerlach, T.: Adjusting Live-Traffic to Match a Given Packet Trace, (2024).
- [ BibTeX ]
- [ EndNote ]
@mastersthesis{info3-praktikum-2024-2, author = {Gerlach, Tim}, keywords = {ucn}, month = {05}, school = {University of Würzburg}, title = {Adjusting Live-Traffic to Match a Given Packet Trace}, year = 2024 }

%0 Thesis %1 info3-praktikum-2024-2 %A Gerlach, Tim %D 2024 %T Adjusting Live-Traffic to Match a Given Packet Trace
1.
Werner, D.: Simulative Evaluation of (In-)Confident Machine Learning for User-based Active Learning, (2023).
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
Machine learning is becoming more and more prevalent in networking, e.g, network monitoring and managament. Though, expert knowledge is always valuable, sometimes irreplaceable. As the machine learning model may not always be confident, one idea is to relay such inconfident decisions to an expert, i.e., the admin, which can decide what to do with this decision and potentially relabel it, called user-based active learning. The goal of this thesis is to perform a simulative analysis regarding the influence of various parameters, such as the confidence threshold (i.e., when to relay the information) or the processing time and qualification level of the admin (i.e., a fast but not always correct expert, slow but accurate etc.). For more information, read the linked presentation slide and/or message katharina.dietz@uni-wuerzburg.de.

@mastersthesis{info3-bachelor-2022-13, abstract = {Machine learning is becoming more and more prevalent in networking, e.g, network monitoring and managament. Though, expert knowledge is always valuable, sometimes irreplaceable. As the machine learning model may not always be confident, one idea is to relay such inconfident decisions to an expert, i.e., the admin, which can decide what to do with this decision and potentially relabel it, called user-based active learning. The goal of this thesis is to perform a simulative analysis regarding the influence of various parameters, such as the confidence threshold (i.e., when to relay the information) or the processing time and qualification level of the admin (i.e., a fast but not always correct expert, slow but accurate etc.). For more information, read the linked presentation slide and/or message katharina.dietz@uni-wuerzburg.de.}, author = {Werner, David}, keywords = {ucn}, month = 12, school = {University of Würzburg}, title = {Simulative Evaluation of (In-)Confident Machine Learning for User-based Active Learning}, year = 2023 }

%0 Thesis %1 info3-bachelor-2022-13 %A Werner, David %D 2023 %T Simulative Evaluation of (In-)Confident Machine Learning for User-based Active Learning %X Machine learning is becoming more and more prevalent in networking, e.g, network monitoring and managament. Though, expert knowledge is always valuable, sometimes irreplaceable. As the machine learning model may not always be confident, one idea is to relay such inconfident decisions to an expert, i.e., the admin, which can decide what to do with this decision and potentially relabel it, called user-based active learning. The goal of this thesis is to perform a simulative analysis regarding the influence of various parameters, such as the confidence threshold (i.e., when to relay the information) or the processing time and qualification level of the admin (i.e., a fast but not always correct expert, slow but accurate etc.). For more information, read the linked presentation slide and/or message katharina.dietz@uni-wuerzburg.de.
1.
Michler, J.: Correlating Machine Learning Confidence to Outlier Detection for Network Monitoring Tasks, (2023).
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
The detection of outliers in network traffic is an omnipresent topic for network security, e.g., to detect malicious activity or unknown devices and apps in the network, that deviate from the normal network activity. The goal of this thesis is to compare a myriad of different detection algorithms and compare them not only regarding their accuracy and detection time, but also with regards to the false negatives and false positives, i.e., investigate the outliers they disagreed on, which were falsely detected etc., as this may be a reason to include more than one algorithm, e.g., in a monitoring system. For more information, read the linked presentation slide and/or message katharina.dietz@uni-wuerzburg.de.

@mastersthesis{info3-bachelor-2022-16, abstract = {The detection of outliers in network traffic is an omnipresent topic for network security, e.g., to detect malicious activity or unknown devices and apps in the network, that deviate from the normal network activity. The goal of this thesis is to compare a myriad of different detection algorithms and compare them not only regarding their accuracy and detection time, but also with regards to the false negatives and false positives, i.e., investigate the outliers they disagreed on, which were falsely detected etc., as this may be a reason to include more than one algorithm, e.g., in a monitoring system. For more information, read the linked presentation slide and/or message katharina.dietz@uni-wuerzburg.de.}, author = {Michler, Jonas}, keywords = {ucn}, month = 11, school = {University of Würzburg}, title = {Correlating Machine Learning Confidence to Outlier Detection for Network Monitoring Tasks}, year = 2023 }

%0 Thesis %1 info3-bachelor-2022-16 %A Michler, Jonas %D 2023 %T Correlating Machine Learning Confidence to Outlier Detection for Network Monitoring Tasks %X The detection of outliers in network traffic is an omnipresent topic for network security, e.g., to detect malicious activity or unknown devices and apps in the network, that deviate from the normal network activity. The goal of this thesis is to compare a myriad of different detection algorithms and compare them not only regarding their accuracy and detection time, but also with regards to the false negatives and false positives, i.e., investigate the outliers they disagreed on, which were falsely detected etc., as this may be a reason to include more than one algorithm, e.g., in a monitoring system. For more information, read the linked presentation slide and/or message katharina.dietz@uni-wuerzburg.de.
1.
Glauer, S.: Experimental Evaluation Tool for Exploring User-based Active Learning for ML-based Browser and URL Detection, (2022).
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
While Machine Learning (ML) is gaining more and more popularity, the actual incorporation into running systems is still lacking, e.g., due to missing trust of network operators. To build this trust, the goal of this work is to incorporate an active learning approach into an already existing ML-approach, that opens up the possibility of making ML-based decisions more explainable and allows to incorporate feedback of the administrator, e.g.,to retrain ML models with the adjusted decisions.

@mastersthesis{info3-bachelor-2021-22, abstract = {While Machine Learning (ML) is gaining more and more popularity, the actual incorporation into running systems is still lacking, e.g., due to missing trust of network operators. To build this trust, the goal of this work is to incorporate an active learning approach into an already existing ML-approach, that opens up the possibility of making ML-based decisions more explainable and allows to incorporate feedback of the administrator, e.g.,to retrain ML models with the adjusted decisions.}, author = {Glauer, Steven}, keywords = {ucn}, month = {09}, school = {University of Würzburg}, title = {Experimental Evaluation Tool for Exploring User-based Active Learning for ML-based Browser and URL Detection}, year = 2022 }

%0 Thesis %1 info3-bachelor-2021-22 %A Glauer, Steven %D 2022 %T Experimental Evaluation Tool for Exploring User-based Active Learning for ML-based Browser and URL Detection %X While Machine Learning (ML) is gaining more and more popularity, the actual incorporation into running systems is still lacking, e.g., due to missing trust of network operators. To build this trust, the goal of this work is to incorporate an active learning approach into an already existing ML-approach, that opens up the possibility of making ML-based decisions more explainable and allows to incorporate feedback of the administrator, e.g.,to retrain ML models with the adjusted decisions.
1.
Ceanuri Devesa, J.K.: Performance Evaluation of Feature Representation Transfer Approaches for Video QoE Estimation Across Different Networks, (2022).
- [ BibTeX ]
- [ EndNote ]
@mastersthesis{info3-master-2022-5, author = {Ceanuri Devesa, Jaione Karmele}, keywords = {ucn}, month = {07}, school = {University of Würzburg}, title = {Performance Evaluation of Feature Representation Transfer Approaches for Video QoE Estimation Across Different Networks}, year = 2022 }

%0 Thesis %1 info3-master-2022-5 %A Ceanuri Devesa, Jaione Karmele %D 2022 %T Performance Evaluation of Feature Representation Transfer Approaches for Video QoE Estimation Across Different Networks
1.
Sichermann, M.: Building a GAN from Scratch for Synthesizing Data Samples for the Use Case of Browser Fingerprinting, (2022).
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
With the recent rise of Machine Learning and especially Deep Learning, gathering enough data is a challenging task. Generative Adversarial Networks (GANs) have emerged to synthesize fake data from real data, by employing two neural networks that interact with each other, i.e., a discriminator and a generator, where the generator tries to trick the discriminator into classifying a fake sample as real. GANs are usually employed on images, however, this has recently been adopted in other research areas. This thesis investigates the capability of GANs to generate fake samples of network traffic, e.g., for data augmentation to balance out datasets (for intrusion detection, for example).

@mastersthesis{info3-bachelor-2022-7, abstract = {With the recent rise of Machine Learning and especially Deep Learning, gathering enough data is a challenging task. Generative Adversarial Networks (GANs) have emerged to synthesize fake data from real data, by employing two neural networks that interact with each other, i.e., a discriminator and a generator, where the generator tries to trick the discriminator into classifying a fake sample as real. GANs are usually employed on images, however, this has recently been adopted in other research areas. This thesis investigates the capability of GANs to generate fake samples of network traffic, e.g., for data augmentation to balance out datasets (for intrusion detection, for example).}, author = {Sichermann, Marleen}, keywords = {ucn}, month = {03}, school = {University of Würzburg}, title = {Building a GAN from Scratch for Synthesizing Data Samples for the Use Case of Browser Fingerprinting}, year = 2022 }

%0 Thesis %1 info3-bachelor-2022-7 %A Sichermann, Marleen %D 2022 %T Building a GAN from Scratch for Synthesizing Data Samples for the Use Case of Browser Fingerprinting %X With the recent rise of Machine Learning and especially Deep Learning, gathering enough data is a challenging task. Generative Adversarial Networks (GANs) have emerged to synthesize fake data from real data, by employing two neural networks that interact with each other, i.e., a discriminator and a generator, where the generator tries to trick the discriminator into classifying a fake sample as real. GANs are usually employed on images, however, this has recently been adopted in other research areas. This thesis investigates the capability of GANs to generate fake samples of network traffic, e.g., for data augmentation to balance out datasets (for intrusion detection, for example).
1.
Ebner, M.: Plausible Performance Prediction of Simulated SDN-enabled Networks via Machine Learning, (2021).
- [ Abstract ]
- [ BibTeX ]
- [ EndNote ]
In recent years, Software-Defined Networking (SDN) emerged as an innovative network architecture and gained increasing popularity due to its centralized network management. However, this central point induces a potential bottleneck and possible single-point-of-failure to the network. Therefore, so-called distributed architectures come into play, which keep the logical view centralized, but physically distributed. The goal of this thesis is to evaluate two opposing approaches and predict their performance via Machine Learning (ML). The simulation will be conducted on basis of a preexisting framework based on OMNeT++, which already implements two distributed SDN-architectures, ready for simulation.

@mastersthesis{info3-bachelor-2021-9, abstract = {In recent years, Software-Defined Networking (SDN) emerged as an innovative network architecture and gained increasing popularity due to its centralized network management. However, this central point induces a potential bottleneck and possible single-point-of-failure to the network. Therefore, so-called distributed architectures come into play, which keep the logical view centralized, but physically distributed. The goal of this thesis is to evaluate two opposing approaches and predict their performance via Machine Learning (ML). The simulation will be conducted on basis of a preexisting framework based on OMNeT++, which already implements two distributed SDN-architectures, ready for simulation.}, author = {Ebner, Markus}, keywords = {ucn}, month = {07}, school = {University of Würzburg}, title = {Plausible Performance Prediction of Simulated SDN-enabled Networks via Machine Learning}, year = 2021 }

%0 Thesis %1 info3-bachelor-2021-9 %A Ebner, Markus %D 2021 %T Plausible Performance Prediction of Simulated SDN-enabled Networks via Machine Learning %X In recent years, Software-Defined Networking (SDN) emerged as an innovative network architecture and gained increasing popularity due to its centralized network management. However, this central point induces a potential bottleneck and possible single-point-of-failure to the network. Therefore, so-called distributed architectures come into play, which keep the logical view centralized, but physically distributed. The goal of this thesis is to evaluate two opposing approaches and predict their performance via Machine Learning (ML). The simulation will be conducted on basis of a preexisting framework based on OMNeT++, which already implements two distributed SDN-architectures, ready for simulation.
1.
Khelloqi, S.: Evaluation of Data Augmentation for Machine Learning on Network Traffic, (2021).
- [ BibTeX ]
- [ EndNote ]
@mastersthesis{info3-bachelor-2020-9, author = {Khelloqi, Sanaa}, keywords = {ucn}, month = {04}, school = {University of Würzburg}, title = {Evaluation of Data Augmentation for Machine Learning on Network Traffic}, year = 2021 }

%0 Thesis %1 info3-bachelor-2020-9 %A Khelloqi, Sanaa %D 2021 %T Evaluation of Data Augmentation for Machine Learning on Network Traffic

Katharina Dietz M. Sc.

Lectures (Teaching Assistant)

Seminar Talks (Tutor)

Supervised Theses

Hinweis zum Datenschutz

Hinweis zum Datenschutz

Bildnachweise