Wave Ngampruetikorn

I study learning and intelligence through the lens of physics. My research group blends ideas from statistical physics, information theory and machine learning to understand how intelligent behavior emerges in complex systems like deep neural networks, treating AI systems not as a mere engineering problem but as a new frontier for science. I am particularly interested in the surprising, emergent phenomena of modern AI. Current research themes include the principles behind in-context learning, the universality of neural computation, and the application of physics-motivated concepts like coarse-graining to make AI more interpretable.

I received my PhD in Physics from the University of Cambridge (Theory of Condensed Matter group), followed by postdoctoral research at OIST (Biological Physics Theory Unit), Northwestern University (Condensed Matter Theory group), and CUNY (Initiative for the Theoretical Sciences & Center for the Physics of Biological Function). I am now a Senior Lecturer (Assistant Professor) in the School of Physics at the University of Sydney.

join us! I am looking for motivated and talented PhD students to join the group. If you are passionate about statistical physics, machine learning and understanding the fundamental principles of intelligence (biological or artificial), I encourage you to get in touch.

selected publications

PRE

Data coarse graining can improve model performance

Alex Nguyen, David J Schwab^*, and Vudtiwat Ngampruetikorn^*

Physical Review E, Feb 2026

Featured in Statistical Physics Meets Machine Learning - Machine Learning Meets Statistical Physics collection

Reducing data resolution can improve predictions by filtering low-relevance features

Paper Cite

@article{nguyen2026data,
  title = {Data coarse graining can improve model performance},
  author = {Nguyen, Alex and Schwab, David J and Ngampruetikorn, Vudtiwat},
  year = {2026},
  journal = {Physical Review E},
  doi = {10.1103/xn84-qzsw},
  volume = {113},
  issue = {2},
  pages = {025301},
  numpages = {8},
  month = feb,
  publisher = {American Physical Society},
  url = {https://link.aps.org/doi/10.1103/xn84-qzsw},
  summary = {Reducing data resolution can improve predictions by filtering low-relevance features},
  note = {<b>Featured in <a href="https://journals.aps.org/pre/collections/ml-stat-pre">Statistical Physics Meets Machine Learning - Machine Learning Meets Statistical Physics</a> collection</b>}
}

NeurIPS

Generalization vs Specialization under Concept Shift

Alex Nguyen, David J Schwab^*, and Vudtiwat Ngampruetikorn^*

In Advances in Neural Information Processing Systems, Feb 2025

Distribution shift can drive phase transitions in model performance

OpenReview arXiv Cite

@inproceedings{nguyen2025generalization,
  title = {Generalization vs Specialization under Concept Shift},
  author = {Nguyen, Alex and Schwab, David J and Ngampruetikorn, Vudtiwat},
  booktitle = {Advances in Neural Information Processing Systems},
  year = {2025},
  url = {https://openreview.net/forum?id=kKPP6q5vBs},
  summary = {Distribution shift can drive phase transitions in model performance}
}

ICML

When can in-context learning generalize out of task distribution?

Chase Goddard, Lindsay M Smith, Vudtiwat Ngampruetikorn^*, and David J Schwab^*

In Forty-second International Conference on Machine Learning, Feb 2025

Pretraining task diversity determines whether models generalize or specialize

OpenReview Cite

@inproceedings{goddard2025when,
  title = {When can in-context learning generalize out of task distribution?},
  author = {Goddard, Chase and Smith, Lindsay M and Ngampruetikorn, Vudtiwat and Schwab, David J},
  booktitle = {Forty-second International Conference on Machine Learning},
  year = {2025},
  url = {https://openreview.net/forum?id=YKyza9lrv4},
  summary = {Pretraining task diversity determines whether models generalize or specialize}
}

PRResearch

Extrinsic vs intrinsic criticality in systems with many components

Vudtiwat Ngampruetikorn, Ilya Nemenman^*, and David J Schwab^*

Physical Review Research, Feb 2025

Apparent criticality in nature can emerge from hidden variables without fine-tuning

Paper Cite

@article{PhysRevResearch.7.013188,
  title = {Extrinsic vs intrinsic criticality in systems with many components},
  author = {Ngampruetikorn, Vudtiwat and Nemenman, Ilya and Schwab, David J},
  journal = {Physical Review Research},
  volume = {7},
  issue = {1},
  pages = {013188},
  numpages = {17},
  year = {2025},
  _month = {Feb},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevResearch.7.013188},
  url = {https://link.aps.org/doi/10.1103/PhysRevResearch.7.013188},
  summary = {Apparent criticality in nature can emerge from hidden variables without fine-tuning}
}

PRL

Extreme Synergy in the Random-Energy Model

Vudtiwat Ngampruetikorn and David J Schwab

Physical Review Letters, Feb 2024

Spin glasses naturally encode information like cryptographic secret-sharing schemes

Paper Cite

@article{PhysRevLett.133.257401,
  title = {Extreme Synergy in the Random-Energy Model},
  author = {Ngampruetikorn, Vudtiwat and Schwab, David J},
  journal = {Physical Review Letters},
  volume = {133},
  issue = {25},
  pages = {257401},
  numpages = {7},
  year = {2024},
  _month = {Dec},
  publisher = {American Physical Society},
  doi = {10.1103/PhysRevLett.133.257401},
  url = {https://link.aps.org/doi/10.1103/PhysRevLett.133.257401},
  summary = {Spin glasses naturally encode information like cryptographic secret-sharing schemes}
}

NeurIPS

Information bottleneck theory of high-dimensional regression: relevancy, efficiency and optimality

Vudtiwat Ngampruetikorn and David J Schwab

In Advances in Neural Information Processing Systems, Feb 2022

Spotlight presentation

Information-theoretic framework reveals the tradeoff between overfitting and learning

Paper Cite

@inproceedings{ngampruetikorn2022information,
  author = {Ngampruetikorn, Vudtiwat and Schwab, David J},
  booktitle = {Advances in Neural Information Processing Systems},
  editor = {Koyejo, S. and Mohamed, S. and Agarwal, A. and Belgrave, D. and Cho, K. and Oh, A.},
  pages = {9784--9796},
  publisher = {Curran Associates, Inc.},
  title = {Information bottleneck theory of high-dimensional regression: relevancy, efficiency and optimality},
  url = {https://proceedings.neurips.cc/paper_files/paper/2022/hash/3fbcfbc2b4009ae8dfa17a562532d123-Abstract-Conference.html},
  volume = {35},
  year = {2022},
  note = {<b>Spotlight presentation</b>},
  summary = {Information-theoretic framework reveals the tradeoff between overfitting and learning}
}

View all publications →