Machine Learning Research Unit

Welcome to the web-home of the new Machine Learning research unit of TU Wien Informatics.

Machine Learning Research Unit (E 194-06) Information Systems Engineering Institute Faculty of Informatics TU Wien Web: https://ml-tuw.github.io/ Physical: Erzherzog-Johann-Platz 1 (FB02), 1040 Vienna, Austria

News

1. Two papers at COLT 2025: A fine-grained characterization of PAC learnability and Of dice and games: a theory of generalized boosting. (May '25)
2. Three papers at ICML 2025: Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective, Probably Approximately Global Robustness, and WILTing Trees: Interpreting the Distance Between MPNN Embeddings. (May '25)
3. Francesco Flaviano De Santis joined us as a PhD student. (April '25)
4. Our work on Lifted inference beyond first order logic got accepted at the Artificial Intelligence Journal. (February '25)
5. Guillaume Bellec will join our research group in 2025. He was awarded a Vienna Research Group! (January '25)
6. We'll present Self-Directed Learning of Convex Labelings on Graphs at ALT'2025. (Dezember '24)
7. Simple and Effective Transfer Learning for Neuro-Symbolic Integration got the Best Paper Award at NeSy'2024. (November '24)
8. Klaus Weinbauer joined our group as a PhD student. (November '24)
9. Logical Distillation of Graph Neural Networks got an Honorable Mention Award as the best paper in the Reasoning, Learning, and Decision Making Track at KR'2024. (October'24)
10. A warm welcome to Benoît Goupil (CentraleSupélec) who visits our lab for six months! (October '24)
11. We'll present two papers at NeurIPS: Weisfeiler and Leman Go Loopy: A New Hierarchy for Graph Representational Learning (oral) and Bandits with Abstention under Expert Advice. (October '24)
12. Splitting Stump Forests received the Best Student Paper Award at the International Conference on Discovery Science. (October '24)
13. Our paper Simple and Effective Transfer Learning for Neuro-Symbolic Integration has been accepted at the NeSy conference as a Spotlight! (August '24)
14. Logical classifiers on graphs that look very similar to MPNNs? Yes! Check out our paper that got recently accepted to KR'2024. (August '24)
15. Are your random forests too large, as well? Try Splitting Stump Forests to compress them without losing accuracy! Freshly accepted at DS'2024 (August '24)
16. Can you verify a smaller distilled student model to prove robustness of a larger teacher model? Yes! With PAC guarantees! Checkout out our paper Distillation based Robustness Verification with PAC Guarantees that got accepted at the ICML workshop for Next Generation AI Safety (July '24)
17. We have organized the Graph ML social at ICML'24 in beautiful Vienna. It was a great success with panel discussions, food, drinks, and over 200 participants (July '24)
18. Johannes Petersen joined as the newest member of our research group! (July '24)
19. We'll present two papers at ECML/PKDD: Understanding Domain-Size Generalization in Markov Logic Networks in the research track, and Scalable Interactive Data Visualization in the demo track. (June '24)
20. A new publication in the Journal of Cheminformatics: Reaction Rebalancing: A Novel Approach to Curating Reaction Databases (June '24)

Old News

1. Two accepted papers at COLT'24! Efficient Algorithms for Learning Monophonic Halfspaces in Graphs and A Theory of Interpretable Approximations. (May '24)
2. Our paper The Expressive Power of Path based Graph Neural Networks has been accepted at ICML'24. (May '24)
3. Dr. Fabrizio Frasca (Technion) and Masahiro Negishi (University of Tokyo) are visiting our lab for the summer term. A warm welcome to them! (March '24)
4. We are excited to announce that we are organizing the 22nd International Workshop on Mining and Learning with Graphs (MLG). It will again be held jointly with ECML/PKDD in Vilnius. (March '24)
5. We hired Christoph Sandrock as newest member of our research group! (March '24)
6. Stefan Neumann received funding for a Vienna Research Group and will join our research unit in 2024. (October '23)
7. Our paper Expressivity-Preserving GNN Simulation was accepted at NeurIPS'23. (October '23)
8. We won the best poster award at G-Research's ICML poster party in London where our very own Maximilian Thiessen presented Expectation-Complete Graph Representations with Homomorphisms.
9. We hired two amazing new PostDocs! Dr. Pascal Welke on our StruDL project and Dr. Sagar Malhotra . Welcome to the team! (May '23)
10. Our paper Deep Symbolic Learning: Discovering Symbols and Rules from Perceptions has been accepted at IJCAI'23. It has also been accepted for spotlight presentation at the NeSy'23 workshop. (May '23)
11. Our paper An Empirical Evaluation of the Rashomon Effect in Explainable Machine Learning got accepted at ECMLPKDD'23. (May '23)
12. Two accepted papers at ACL'23! Hidden Schema Networks and A New Aligned Simple German Corpus. (May '23)
13. Our paper on Expectation-Complete Graph Representations with Homomorphisms has been accepted at ICML'23! (April '23)
14. Our paper Can stochastic weight averaging improve generalization in private learning? has been accepted at the RTML workshop of ICLR'23! (April '23)
15. We are organising the Mining and Learning with Graphs (MLG) workshop this year at ECMLPKDD2023. (April '23)
16. We organised the 1st community event for students of learning algorithms in Vienna (C'est La Wien, Feb '23)
17. Caterina Graziani (Università degli Studi di Siena) is visiting us for four months! (Feb '23)
18. Our paper on Krein support vector machine classification of antimicrobial peptides has been accepted for publication in the Digital Discovery journal! Joint work of Joseph Redshaw, Darren S. J. Ting, Alex Brown, Jonathan D. Hirst, and Thomas Gärtner . (Feb '23)
19. Two of our papers have been accepted at NeurIPS'22! (Oct '22)
20. Our very own Fabian Jogl won the (community voted) Best Poster Award at the MLG@ECMLPKDD Workshop. Congratulations! (Sep '22)
21. We organised machine learning courses for children at the KinderUni Wien. (Jul '22)
22. Together with Pascal Welke from the University of Bonn, we are organising the Mining and Learning with Graphs (MLG) workshop this year at ECMLPKDD'22. Submit your work until June 20th!
23. Our very own Tamara Drucks won the Best Poster Award of TU Wien Informatics. Congratulations! (Jan '22)
    
24. Our paper on Kernel Methods for Predicting Yields of Chemical Reactions has been accepted for publication in the Journal of Chemical Information and Modeling. Joint work of Alexe Haywood, Joseph Redshaw, Magnus Hanson-Heine, Adam Taylor, Alex Brown, Andrew Mason, Thomas Gärtner , and Jonathan Hirst. (Oct '21)
25. Our paper on Conditional Network Data Balancing With GANs has been accepted at the NeurIPS Deep Generative Models and Downstream Applications Workshop. Joint work ok Fares Meghdouri, Thomas Schmied, Tanja Zseby, and Thomas Gärtner . (Oct '21)
26. Our paper on Active Learning Convex Halfspaces on Graphs (supplementary material) has been accepted at NeurIPS'21. Congratulations to Maximilian Thiessen for his first NeurIPS paper! (Sept '21)
    
27. Our team won the prize for the Best Practical Result in the Siemens AI-Dependability Assessment Challenge. Congratulations to Joseph Redshaw (University of Nottingham), Maximilian Thiessen (TU Wien), and David Penz (TU Wien)! (July '21)
    
28. Our very own David Penz won the Distinguished Young Alumn award of TU Wien Informatics. Congratulations! (June '21)
    
29. FFG has aggreed to fund our proposal on Artificial Intelligence for Advanced SAR Processing! (May '21)
30. TU Wien has agreed to fund our doctoral college on Secure and Intelligent Human-Centric Digital Technologies ! (July '20)
31. Founding of the Machine Learning research unit. (Feb '20)
32. BBSRC has aggreed to fund our proposal on Bacteriophage to control Salmonella in pigs! (Oct '19)

Topics

Our research aims to narrow the gap between theoretically well-understood and practically relevant machine learning. Research questions concern for instance:

• learning with non-conventional data, i.e., data that has no inherent representation in a table or Euclidean space
• incorporation of invariances as well as expert domain knowledge in learning algorithms
• computational, sample, query, and communication complexity of learning algorithms
• constructive machine learning scenarios such as structured output prediction
• learning with small labelled data sets and large unlabelled data sets
• adversarial learning with mistake and/or regret bounds
• parallelisation/distribution of learning algorithms
• approximation of learning algorithms
• scalability of learning algorithms
• reliability of learning algorithms
• extreme learning
• ...

To demonstrate the practical effectiveness of novel learning algorithms, we apply them in Chemistry, Material Science, Electrical Engineering, Computer Games, Humanities, etc.