Semih Cantürk
PhD candidate at Mila and Université de Montréal, in Guy Wolf’s group.
About
My research spans the theory and applications of geometric and topological deep learning, with a focus on graph representation learning and spectral graph theory.
I’m particularly interested in solving combinatorial optimization problems with graph learning, in geometric learning on biomolecular data, and in positional and structural encodings that make graph learning more effective and scalable. I work in Guy Wolf’s group at Mila – Quebec AI Institute and Université de Montréal (DIRO).
I did my MSc at Mila and UdeM as well, and my BSE at the University of Pennsylvania. In fall 2025 I was a visiting researcher in Christopher Morris’s LOG group at RWTH Aachen, and in 2024 a PhD intern at Valence Labs, working on accelerating molecular dynamics simulations with machine-learned interatomic potentials. Before my PhD I was an ML engineer at Zetane Systems.
I speak Turkish, English, conversational Spanish, and some Greek and French. Away from research: tennis, football, skiing, guitar, literature, history and anthropology.
Education
MSc in Computer Science
Mila & Université de Montréal
Advised by Guy Wolf. Thesis: Taxonomy of Datasets in Graph Learning: A Data-Driven Approach to Improve GNN Benchmarking.
BSE in Systems Science & Engineering
University of Pennsylvania
Minors in computer science and mathematics, magna cum laude. Thesis: Motor Task Prediction through fMRI Data, advised by Robert Ghrist.
News
New preprint with Lydia Mezrag: Path Laplacian Encodings for Directed Graphs, a spectral positional encoding that keeps edge direction.
At ICML 2026 in Seoul: Can Computational Reducibility Lead to Transferable Models for Graph Combinatorial Optimization? in the main conference, and a position paper on neural algorithmic reasoning at the AI for Math workshop.
New preprint: GraIP, a benchmarking framework that treats structure discovery and other graph learning tasks as inverse problems.
Gave a tutorial on geometric and topological representation learning at IEEE MLSP 2025 in Istanbul.
Started a fall visit to Christopher Morris’s LOG group at RWTH Aachen.
Earlier news (8)
Towards Graph Foundation Models was accepted to TMLR, with a Reproducibility Certification.
Towards a General Recipe for Combinatorial Optimization with Multi-Filter GNNs was accepted to LoG 2024 as a Spotlight.
Two new preprints: Towards Graph Foundation Models, a follow-up to GPSE, and OpenQDC, a large collection of open-source quantum molecular datasets developed at Valence Labs.
GPSE was accepted to ICML 2024.
Awarded the Université de Montréal PhD Scholarship in Artificial Intelligence (Bourse en intelligence artificielle des ESP).
Our preprint Graph Positional and Structural Encoder (GPSE) is out on arXiv.
My MSc thesis is published.
Submitted my MSc thesis and started my PhD at Mila and UdeM in Guy Wolf’s group.
Papers
* equal contribution
Path Laplacian Encodings for Directed Graphs
@misc{mezrag2026pathlaplacian, title = {Path Laplacian Encodings for Directed Graphs}, author = {Lydia Mezrag and Semih Cant{\"u}rk and Michael Perlmutter and Bastian Rieck and Guy Wolf}, year = {2026}, eprint = {2610.04657}, archivePrefix = {arXiv}, primaryClass = {cs.LG}, url = {https://arxiv.org/abs/2610.04657}, }Can Computational Reducibility Lead to Transferable Models for Graph Combinatorial Optimization?
@inproceedings{canturk2026reducibility, title = {Can Computational Reducibility Lead to Transferable Models for Graph Combinatorial Optimization?}, author = {Semih Cant{\"u}rk and Thomas Sabourin and Frederik Wenkel and Michael Perlmutter and Guy Wolf}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning (ICML)}, year = {2026}, url = {https://arxiv.org/abs/2603.02462}, }Neural Algorithmic Reasoning Must Explain When Neuralization Adds Value
@inproceedings{he2026nar, title = {Neural Algorithmic Reasoning Must Explain When Neuralization Adds Value}, author = {Yu He and Robert R. Nerem and Timo Stoll and Semih Cant{\"u}rk and Dobrik Georgiev and Solveig Wittig and Chendi Qian and Floris Geerts and Stefanie Jegelka and Ellen Vitercik and Yusu Wang and Nikolaos Karalias and Christopher Morris}, booktitle = {3rd AI for Math Workshop at ICML 2026}, year = {2026}, url = {https://openreview.net/forum?id=A4FI5tZRT4}, }GraIP: A Benchmarking Framework For Neural Graph Inverse Problems
@misc{canturk2026graip, title = {GraIP: A Benchmarking Framework For Neural Graph Inverse Problems}, author = {Semih Cant{\"u}rk and Arman Mielke and Andrei Manolache and Chendi Qian and Antoine Siraudin and Christopher Morris and Mathias Niepert and Guy Wolf}, year = {2026}, eprint = {2601.18917}, archivePrefix = {arXiv}, primaryClass = {cs.LG}, url = {https://arxiv.org/abs/2601.18917}, }Towards Graph Foundation Models: A Study on the Generalization of Positional and Structural Encodings
@article{franks2025graphfoundationmodels, title = {Towards Graph Foundation Models: A Study on the Generalization of Positional and Structural Encodings}, author = {Billy Joe Franks and Moshe Eliasof and Semih Cant{\"u}rk and Guy Wolf and Carola-Bibiane Sch{\"o}nlieb and Sophie Fellenz and Marius Kloft}, journal = {Transactions on Machine Learning Research}, issn = {2835-8856}, year = {2025}, url = {https://openreview.net/forum?id=mSoDRZXsqj}, note = {Reproducibility Certification}, }OpenQDC: Open Quantum Data Commons
@misc{gabellini2024openqdc, title = {OpenQDC: Open Quantum Data Commons}, author = {Cristian Gabellini and Nikhil Shenoy and Stephan Thaler and Semih Cant{\"u}rk and Daniel McNeela and Dominique Beaini and Michael Bronstein and Prudencio Tossou}, year = {2024}, eprint = {2411.19629}, archivePrefix = {arXiv}, primaryClass = {physics.chem-ph}, url = {https://arxiv.org/abs/2411.19629}, }Towards a General Recipe for Combinatorial Optimization with Multi-Filter GNNs
@inproceedings{wenkel2024general, title = {Towards a General Recipe for Combinatorial Optimization with Multi-Filter GNNs}, author = {Frederik Wenkel and Semih Cant{\"u}rk and Stefan Horoi and Michael Perlmutter and Guy Wolf}, booktitle = {Proceedings of the Third Learning on Graphs Conference (LoG)}, series = {Proceedings of Machine Learning Research}, volume = {269}, publisher = {PMLR}, year = {2024}, url = {https://arxiv.org/abs/2405.20543}, }Graph Positional and Structural Encoder
@inproceedings{canturk2024gpse, title = {Graph Positional and Structural Encoder}, author = {Semih Cant{\"u}rk and Renming Liu and Olivier Lapointe-Gagn{\'e} and Vincent L{\'e}tourneau and Guy Wolf and Dominique Beaini and Ladislav Ramp{\'a}{\v{s}}ek}, booktitle = {Proceedings of the 41st International Conference on Machine Learning (ICML)}, series = {Proceedings of Machine Learning Research}, volume = {235}, publisher = {PMLR}, year = {2024}, url = {https://arxiv.org/abs/2307.07107}, }Taxonomy of Benchmarks in Graph Representation Learning
@inproceedings{liu2022taxonomy, title = {Taxonomy of Benchmarks in Graph Representation Learning}, author = {Liu, Renming and Cant{\"u}rk, Semih and Wenkel, Frederik and McGuire, Sarah and Wang, Xinyi and Little, Anna and O'Bray, Leslie and Perlmutter, Michael and Rieck, Bastian and Hirn, Matthew and Wolf, Guy and Ramp{\'a}{\v{s}}ek, Ladislav}, booktitle = {Proceedings of the First Learning on Graphs Conference (LoG)}, series = {Proceedings of Machine Learning Research}, volume = {198}, pages = {6:1--6:25}, editor = {Rieck, Bastian and Pascanu, Razvan}, publisher = {PMLR}, year = {2022}, url = {https://proceedings.mlr.press/v198/liu22a.html}, }Towards a Taxonomy of Graph Learning Datasets
@article{liu2021towards, title = {Towards a Taxonomy of Graph Learning Datasets}, author = {Liu, Renming and Cant{\"u}rk, Semih and Wenkel, Frederik and Sandfelder, Dylan and Kreuzer, Devin and Little, Anna and McGuire, Sarah and O'Bray, Leslie and Perlmutter, Michael and Rieck, Bastian and Hirn, Matthew and Wolf, Guy and Ramp{\'a}{\v{s}}ek, Ladislav}, journal = {Data-Centric AI Workshop, NeurIPS}, year = {2021}, url = {https://arxiv.org/abs/2110.14809}, }Machine-Learning Driven Drug Repurposing for COVID-19
@misc{canturk2020drug, title = {Machine-Learning Driven Drug Repurposing for COVID-19}, author = {Semih Cant{\"u}rk and Aman Singh and Patrick St-Amant and Jason Behrmann}, year = {2020}, eprint = {2006.14707}, archivePrefix = {arXiv}, primaryClass = {cs.LG}, url = {https://arxiv.org/abs/2006.14707}, }
Experience
LOG group, RWTH Aachen
Visiting researcher, Aachen
Visited Christopher Morris’s group to work on graph representation learning.
Valence Labs
PhD intern, Physical Simulations, Montréal
Built an equivariant GNN delta-learning framework that exploits the correlation between successive states of molecular dynamics simulations, so smaller interatomic potentials can be used without losing accuracy. Contributed experiments and visualizations to OpenQDC, an open library of quantum-mechanical datasets.
Zetane Systems
Researcher and ML software developer, Montréal
Developed the explainability module of the Zetane Engine for computer vision (class-activation maps, SHAP, LIME), and the data-augmentation and explainability modules of Zetane Protector. Led and supported ML projects with industry partners in robotics, energy, construction and automotive.
University of Pennsylvania
Undergraduate researcher, with Victor Preciado and Cassiano Becker, Philadelphia
Built a pipeline that predicts motor tasks from fMRI data with signal processing and LSTMs, which won the Penn Engineering Societal Impact Award; later extended it to mesh-based learning and GNNs.
Earlier
InfoTRON, Software development intern, Istanbul. Recognizing and classifying CAD models in AR/VR with ARToolkit and OpenCV.
Imperial College London, Undergraduate researcher, London. A distributed system to run acute3D in the Imperial College Data Observatory.
SAS Analytics, Data scientist intern, Istanbul. Fraud detection with global partners in the insurance sector.
Talks
Geometric and Topological Representation Learning
Tutorial, IEEE International Workshop on Machine Learning for Signal Processing (MLSP), Istanbul
Towards a General Recipe for Combinatorial Optimization with Multi-Filter GNNs
LoG 2024 Montréal Meetup, Mila
Graph Representation Learning: A Gentle Introduction and New Perspectives
Université de Montréal (Sep 2024) and Concordia University (Mar 2024)
Graph Positional and Structural Encoder
LoG 2023 Montréal Meetup, Mila
Graph Representation Learning
Université de Montréal (Nov 2023) and Concordia University (Mar 2023)
Teaching
Spectral Graph Theory MAT 6495
Geometric Data Analysis MAT 6493
Introduction to Dynamical Systems ESE 210
Service
Organizing
Graph Signal Processing (GSP) Workshop, 8th edition
Learning on Graphs (LoG) Montréal meetup
Reviewing
International Conference on Machine Learning (ICML), Gold reviewer
International Conference on Learning Representations (ICLR), Top reviewer in 2026
Learning on Graphs Conference (LoG)
IEEE Transactions on Neural Networks and Learning Systems (TNNLS)
NeurIPS Workshop on Symmetry and Geometry in Neural Representations (NeurReps)
Workshop on Topology, Algebra and Geometry in Data Science (TAG-DS)
ICLR Workshop on Geometrical and Topological Representation Learning
Leadership
Chair and treasurer of the Penn Preceptorials Committee, a student-run seminar series with Penn faculty
Awards
End-of-doctorate scholarship (Bourse de fin d’études doctorales)Université de Montréal
Gold reviewer awardInternational Conference on Machine Learning (ICML)
Top reviewer awardInternational Conference on Learning Representations (ICLR)
Mitacs Globalink Research AwardMitacs
Bourse d’excellence du DIROUniversité de Montréal
PhD Scholarship in Artificial Intelligence (Bourse en intelligence artificielle des ESP)Université de Montréal
Exemption scholarships for the PhD (2022–present) and MSc (2020–2022)Université de Montréal
Societal Impact AwardUniversity of Pennsylvania, Electrical & Systems Engineering
Top 25 College Graduates of 2018, TurkeyStudy in America
Research stipendImperial College Data Science Institute