Piotr Wyrwiński

Poznan, Poland

I am a PhD student in artificial intelligence at Poznan University of Technology, advised by Prof. Krzysztof Krawiec. I also work as a machine learning researcher at the Poznan Supercomputing and Networking Center.

My research asks how neural networks can learn representations with explicit structure and use them for reasoning or search. I work on learned discrete representations and program synthesis, including neural guidance for symbolic regression. At PSNC, I also work on regional weather forecasting with foundation models.

In STROP, we learn variable-length visual programs: sequences of discrete codes whose active length adapts to each image. A direction I am exploring is compositionality: whether these codes preserve reusable structure and generalise to unfamiliar combinations.

If you’re a student interested in structured representations or neurosymbolic learning, feel free to contact me or Prof. Krzysztof Krawiec about possible research projects.

news

Sep 30, 2026	An updated version of our STROP paper, Structure over Pixels: Learning Variable-Length Visual Programs, is now available on arXiv (v3).
Sep 30, 2026	I’ll present updated STROP work at PTA: From Pretrained Representations to Acting Agents, a NeurIPS 2026 workshop in Sydney on 12 December.
Aug 01, 2026	I attended the Eastern European Machine Learning Summer School (EEML) in Cetinje, Montenegro, from 27 July to 1 August.
Jul 11, 2026	I presented STROP at CompLearn: Compositional Learning, Safety, Interpretability, and Agents, a workshop at ICML 2026 in Seoul.

selected publications

Preprint

Structure over Pixels: Learning Variable-Length Visual Programs

Piotr Wyrwiński, Kacper Dobek, and Krzysztof Krawiec

arXiv preprint arXiv:2605.27696, 2026

Presented at CompLearn, ICML 2026

arXiv Bib HTML PDF

@article{wyrwinski2026strop,
  title = {Structure over Pixels: Learning Variable-Length Visual Programs},
  author = {Wyrwiński, Piotr and Dobek, Kacper and Krawiec, Krzysztof},
  journal = {arXiv preprint arXiv:2605.27696},
  year = {2026},
  note = {Presented at CompLearn, ICML 2026},
}

GECCO

Learning Semantics-aware Search Operators for Genetic Programming

Piotr Wyrwiński and Krzysztof Krawiec

In Proceedings of the Genetic and Evolutionary Computation Conference Companion, 2025

arXiv Bib HTML PDF

@inproceedings{wyrwinski2025semantics,
  title = {Learning Semantics-aware Search Operators for Genetic Programming},
  author = {Wyrwiński, Piotr and Krawiec, Krzysztof},
  booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference Companion},
  pages = {659--662},
  year = {2025},
  publisher = {Association for Computing Machinery},
}

AAAI

Neuro-Guided Graph Search for Symbolic Regression (Student Abstract)

Piotr Wyrwiński and Krzysztof Krawiec

In Proceedings of the AAAI Conference on Artificial Intelligence, 2025

DOI Bib HTML PDF

@inproceedings{wyrwinski2025nudge,
  title = {Neuro-Guided Graph Search for Symbolic Regression (Student Abstract)},
  author = {Wyrwiński, Piotr and Krawiec, Krzysztof},
  booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
  volume = {39},
  number = {28},
  pages = {29529--29531},
  year = {2025},
  doi = {10.1609/aaai.v39i28.35314},
}

GECCO

Guiding Genetic Programming with Graph Neural Networks

Piotr Wyrwiński and Krzysztof Krawiec

In Proceedings of the Genetic and Evolutionary Computation Conference Companion, 2024

DOI arXiv Bib HTML PDF

@inproceedings{wyrwinski2024guiding,
  title = {Guiding Genetic Programming with Graph Neural Networks},
  author = {Wyrwiński, Piotr and Krawiec, Krzysztof},
  booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference Companion},
  pages = {551--554},
  year = {2024},
  publisher = {Association for Computing Machinery},
  doi = {10.1145/3638530.3654277},
}