This project develops Graph Machine Learning models to predict Kamlet-Taft solvation parameters of ionic liquids, which describe key solvent properties such as polarity and hydrogen bonding ability. Ionic liquids are represented as molecular graphs, and Graph Neural Networks are used to learn structural embeddings, combined with quantum chemical descriptors, to build predictive and interpretable models. The goal is to link molecular structure to solvation behavior and enable prediction for new, previously uncharacterized ionic liquids.

Requirements: Strong interest in graph machine learning with application to the chemical domain

Contact: Nils Kriege

Time-series data is a collection of data points recorded over time, each associated with a specific timestamp. This form of data is prevalent in various fields such as finance, economics, meteorology, healthcare, energy, telecommunications, and transportation. Current algorithms assume that we only have time-series data of the same scaling, but in real-world data time-series often consists of different scalings, e.g. hourly, daily, or weekly weather forecasts.

This project will mainly focus on the development of a clustering algorithm that can handle time series with different scalings. Students working on this project need solid programming skills in Python. More details to the interested student will be provided on request.

Contact: Yllka Velaj

Time-series data is a collection of data points recorded over time, each associated with a specific timestamp. This form of data is prevalent in various fields such as finance, economics, meteorology, healthcare, energy, telecommunications, and transportation. Often we can gather additional information about these time series, which could improve cluster performance. This kind of clustering is also called "multi-view clustering". Unfortunately, currently, there are no algorithms for multi-view clustering of time series.

This project will mainly focus on the development of a clustering algorithm that can use additional information from an original time series to retrieve a better clustering. Students working on this project need solid programming skills in Python. More details to the interested student will be provided on request.

Contact: Yllka Velaj

The process of automatically learning features is termed representation learning and offers the advantage of eliminating the need to predetermine which features are crucial, often saving considerable effort in practical applications. Deep Clustering is a research domain that seeks to apply the successful principles of deep learning to the field of clustering. Deep clustering algorithms can autonomously extract crucial features. Additionally, deep clustering demonstrates scalability to large, high-dimensional datasets, resulting in more precise cluster assignments and predictions for new, previously unseen data points. Unfortunately, the usage of deep representation learning and deep clustering algorithms on time-series data is currently very limited.

This project will mainly focus on the development of a deep clustering algorithm for time-series data. Students working on this project need solid programming skills in Python. More details to the interested student will be provided on request.

Contact: Yllka Velaj

The same data set can often be clustered in more than one meaningful way - e.g., weekly commuters vs. home-office workers or early birds vs. night owls. Both are valid clusterings considering the work situation on the one hand and the lifestyle on the other. Forcing the data into one clustering can lead to mixed and less interpretable results. Contrarily, alternative clusterings are easier to interpret as their results focus on one aspect intrinsic to data. Currently, there is no work known to us that clusters time-series data into multiple alternative meaningful clusters. The algorithms have to be adapted for this kind of data, i.e., including the time information for clustering.

Students working on this project need solid programming skills in Python. More details to the interested student will be provided on request.

Contact: Yllka Velaj

The Voter model is a well-studied stochastic process that models the invasion of a novel trait A (e.g., a new opinion, social meme, genetic mutation, magnetic spin) in a network of individuals (agents, people, genes, particles) carrying an existing resident trait B. Individuals change traits by occasionally sampling the trait of a neighbor, while an invasion bias expresses the stochastic preference to adopt the novel trait A over the resident trait B. The strength of an invasion is measured by the probability that eventually the whole population adopts trait A, i.e., the fixation probability.

The goal is to try different techniques to add edges to the graph and see which one improves the fixation probability. Students working on this project need solid programming skills in Python. More details to the interested student will be provided on request.

Contact: Yllka Velaj

With a recent increase in concern for privacy and data protection, several regulations have been passed to ensure the secureness of personal information in cyberspace. Notably, the right to be forgotten allows any individual to request their presence on a system to be completely erased. Machine learning models trained on data retrieved from such a system also need to be modified to comply with the law. In addition to legal obligations, there is a need to remove the effect of outdated or malicious data to enhance model robustness. This task of eliminating the impact of certain training instances is referred to as Machine Unlearning. A simple yet legitimate solution is to train the model from scratch with the unwanted instances deleted from the training set. However, in a production environment, this naive approach can be too costly in terms of deployment time and computational resources.

Current papers for unlearning do not do a good job on borrowing previous methods and testing them. We can explore how the aggregation methods of different partitions play a role in unlearning in GNN. We can also do an elaborated evaluation on the security aspect with different types of attacks.

Students working on this project need solid programming skills in Python. More details to the interested student will be provided on request.

Contact: Yllka Velaj

Graph neural networks (GNNs) based methods have achieved impressive performance on the node clustering task. However, they are designed on the homophilic assumption of the graph, and clustering on the heterophilic graph is overlooked. Hence, clustering on real-world graphs with various levels of homophily poses a new challenge to the graph research community. For example, previous works either preserve structural information by ignoring the impact of heterophily in the graph structure or only focus on node-level consistency by ignoring class-level consistency.

This project invites bachelor's and master's students to explore heterophily in multi-relational graphs (multiplex networks). Specifically, the goal is to construct the multi-hop relationships of selected layer(s) - relation type(s) - between every node to improve homophily and reduce the impact of heterophily in the graph structure. The initial idea is outlined, and students will be responsible for conducting experiments on published approaches and evaluating them on real-world datasets.

Ideal candidates should have a foundation in machine learning principles, proficiency in Python, and basic knowledge of graph neural networks or deep learning.

[1] Yudi Huang, Ci Nie, Hongqing He, Yujie Mo, Yonghua Zhu, Guoqiu Wen, and Xiaofeng Zhu. Multiplex Graph Representation Learning with Homophily and Consistency. AAAI 2025.

[2] Erlin Pan and Zhao Kang. Beyond Homophily: Reconstructing Structure for Graph-agnostic Clustering. ICML 2023.

[3] Sitao Luan et al. The Heterophilic Graph Learning Handbook: Benchmarks, Models, Theoretical Analysis, Applications and Challenges. ArXiv 2024.

Supervisors: Ylli Sadikaj, Claudia Plant

Contact: Ylli Sadikaj

Optical inspection through visual anomaly detection and segmentation is essential for identifying defective products and precisely locating anomalous regions. Recent methods, such as CLIP [1] and DINO [2], leverage prior knowledge to enhance generalization across a wide range of products. For instance, CLIP-based approaches like [3] adapt its inherent knowledge by defining text prompts for normal and abnormal states, which typically results in binary anomaly classification.

This project invites master students to extend these techniques by developing a method for multi-class anomaly detection. The goal is to design and implement a system that learns text prompt embeddings to better align with image embeddings, thereby accurately detecting and differentiating multiple types of anomalies. The initial concept is outlined, and students will be responsible for its implementation and evaluation on real-world datasets.

Ideal candidates should have a strong foundation in machine learning principles, proficiency in Python, and basic knowledge with deep learning frameworks.

[1] Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. In ICML, pages 8748–8763. PMLR, 2021.

[2] Mathilde Caron, Hugo Touvron, Ishan Misra, Herve Jegou, Julien Mairal, Piotr Bojanowski, and Armand ´ Joulin. Emerging properties in self-supervised vision transformers. In ICCV, pages 9630–9640. IEEE, 2021.

[3] Qihang Zhou, Guansong Pang, Yu Tian, Shibo He, and Jiming Chen. Anomalyclip: Object-agnostic prompt learning for zero-shot anomaly detection. In ICLR. 2024.

Supervisors: Ylli Sadikaj, Claudia Plant

Contact: Ylli Sadikaj

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Michael Wiegand

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Vanja Karan

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Yuxi Xia

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Loris Schoenegger

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Timour Igamberdiev

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Benjamin Roth

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Vasiliki Kougia

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Loris Schoenegger

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Vanja Karan

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Timour Igamberdiev

Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.

Contact: Pedro Henrique Luz de Araujo

The graph edit distance (GED) quantifies the dissimilarity of two graphs as the minimum cost of a sequence of edit operations turning one graph into the other. The problem complexity depends on the choice of the edit operation costs and the properties of the considered graphs. It has been shown that the classical NP-hard subgraph isomorphism problem (SI) and the maximum common subgraph problem (MCS) can be reduced to computing the GED. As a consequence, computing the GED also is NP-hard. However, for SI and MCS a fine distinction of the complexity depending on parameters such as treewidth, degree, and their combination is known. The goal of this project is to investigate the complexity of computing the graph edit distance in restricted graph classes identifying polynomial-time solvable and NP-hard cases.

Requirements: Strong interest in graph algorithms and theoretical computer science

Contact: Nils Kriege

Uncertain graphs represent an important data model for real-world networks, where one can assign an independent probability of existence on every edge. Uncertainty, represented by the edge probability, may arise due to several reasons, e.g., errors in measurements, data integration from inconsistent and ambiguous sources, lack of precise information needs, inference and prediction models, and explicit manipulation for privacy purposes. Due to their rich expressiveness and utility in a wide range of applications, uncertain graphs have prompted a great deal of research by the data mining research community.

This work will mainly focus on designing efficient algorithms for selecting and modifying the probability of a given number of edges in an uncertain graphs. The developed algorithm will be evaluated against other existing methods. An experimental study will be conducted on real-world datasets. Students working on this project need basic knowledge in graph theory and algorithms, and solid programming skills in Python, C/C++ or Java.

Contact: Yllka Velaj, Claudia Plant

For two given graphs, G and H, the maximum common subgraph problem (MCS) asks for the largest graph contained in both G and H. An important application occurs in cheminformatics, where the similarity of molecular graphs needs to be quantified. Unfortunately, MCS is NP-hard unless additional constraints regarding the properties of G, H, and the subgraph are enforced. A polynomial-time solvable case requires that all these graphs are trees. Known negative complexity results for classes of tree-like graphs set narrow boundaries for generalization.

The project aims to investigate particular tree-like graph classes to design, implement, and experimentally evaluate new polynomial-time algorithms. New NP-hardness proofs for specific cases should refine the complexity status further.

Requirements: Strong interest in algorithms and graph theory

Contact: Nils Kriege