London · 51°30′N 000°07′W

Benedikt Brückner

AI research scientist working on formal verification and certified robustness of neural networks.

Benedikt Brückner

About

Now · September 2026

Just back from ECCV in Malmö, where I presented IoUCert. A great week, and a city well worth the trip. Otherwise I've finished my PhD at Imperial and continue to work on topics around robustness verification for perception models and validation for AI agents. Currently most interested in scaling verification to production-size models and exploring which ideas from verification of vision models transfer to LLMs and AI agents.

I'm an AI research scientist at Safe Intelligence in London, working on formal verification and certified robustness of neural networks and AI agents. Most of what I do is about getting verification out of the paper and into contact with real models: verifiers for object detectors, certified training methods, and end-to-end robustness assessments for customers.

I finished my PhD in Computer Science at Imperial College London in 2026, supervised by Alessio Lomuscio, on the formal verification and certified training of neural networks. Before that I studied Industrial Engineering and Management at KIT in Karlsruhe, with a semester abroad in Glasgow, and drifted towards machine learning somewhere in the middle of it.

Away from the desk I sail, ride mountain bikes and play tennis.

Work

Dec 2023 —
AI Research Scientist
Safe Intelligence, London
Verification engines for perception models, certified training, and robustness assessments for clients.
2019 – 2020
Student Research Assistant
KIT Institute of Operations Research, Karlsruhe
Sequential decision making under system-inherent uncertainty.

Education

2021 – 2026
PhD, Computer Science
Imperial College London
Thesis: Formal Verification and Certified Training of Neural Networks, supervised by Alessio Lomuscio. Funded by UKRI through the Centre for Doctoral Training in Safe and Trusted AI.
2018 – 2021
MSc, Industrial Engineering and Management
Karlsruhe Institute of Technology
Semester abroad at the University of Glasgow.
2015 – 2018
BSc, Industrial Engineering and Management
Karlsruhe Institute of Technology

Teaching

2022 – 2023
Graduate Teaching Assistant
Imperial College London
Operations Research (two terms) and Computer Vision (one term).
2017 – 2019
Academic Tutor, Mathematics
KIT Institute of Stochastics
Weekly tutorials for groups of around thirty students.

Publications

* equal contribution

2026

IoUCert: Robustness Verification for Anchor-Based Object Detectors

Benedikt Brückner*, Alejandro J. Mercado*, Yanghao Zhang, Panagiotis Kouvaros, Alessio Lomuscio
ECCV 2026
DOIarXivCopied
@inproceedings{brueckner2026ioucert,
  author        = {Benedikt Br{\"u}ckner and Alejandro J. Mercado and Yanghao
                   Zhang and Panagiotis Kouvaros and Alessio Lomuscio},
  title         = {{IoUCert}: Robustness Verification for Anchor-Based Object
                   Detectors},
  booktitle     = {Proceedings of the 19th European Conference on Computer
                   Vision (ECCV 2026)},
  pages         = {364--384},
  publisher     = {Springer},
  year          = {2026},
  doi           = {10.1007/978-3-032-37574-2\_20},
  url           = {https://link.springer.com/chapter/10.1007/978-3-032-37574-2\_20},
  eprint        = {2603.03043},
  archivePrefix = {arXiv},
  isbn          = {978-3-032-37574-2},
}

Certified Training for Convolutional Perturbations

Benedikt Brückner, Alessio Lomuscio
arXiv preprint Preprint
DOIarXivCopied
@misc{brueckner2026convolutional,
  author        = {Benedikt Br{\"u}ckner and Alessio Lomuscio},
  title         = {Certified Training for Convolutional Perturbations},
  howpublished  = {arXiv preprint arXiv:2607.18195},
  year          = {2026},
  doi           = {10.48550/arXiv.2607.18195},
  url           = {https://arxiv.org/abs/2607.18195},
  eprint        = {2607.18195},
  archivePrefix = {arXiv},
}

Certified Background-Invariant Training via Latent Generative Models

Benedikt Brückner*, Harleen Hanspal*, Alessio Lomuscio
AI4AT Workshop, AAAI 2026 Workshop
Copied
@inproceedings{brueckner2026background,
  author        = {Benedikt Br{\"u}ckner and Harleen Hanspal and Alessio
                   Lomuscio},
  title         = {Certified Background-Invariant Training via Latent
                   Generative Models},
  booktitle     = {Proceedings of the AAAI 2026 Workshop on Artificial
                   Intelligence for Air Transportation (AI4AT)},
  year          = {2026},
}
2025

Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels

Benedikt Brückner, Alessio Lomuscio
AAAI 2025
DOIarXivCopied
@inproceedings{brueckner2025kernels,
  author        = {Benedikt Br{\"u}ckner and Alessio Lomuscio},
  editor        = {Toby Walsh and Julie Shah and Zico Kolter},
  title         = {Verification of Neural Networks Against Convolutional
                   Perturbations via Parameterised Kernels},
  booktitle     = {Thirty-Ninth {AAAI} Conference on Artificial Intelligence,
                   Thirty-Seventh Conference on Innovative Applications of
                   Artificial Intelligence, Fifteenth Symposium on Educational
                   Advances in Artificial Intelligence, {AAAI} 2025,
                   Philadelphia, PA, USA, February 25 - March 4, 2025},
  pages         = {27215--27223},
  publisher     = {{AAAI} Press},
  year          = {2025},
  doi           = {10.1609/aaai.v39i26.34930},
  url           = {https://doi.org/10.1609/aaai.v39i26.34930},
  eprint        = {2411.04594},
  archivePrefix = {arXiv},
}

Dynamic Back-Substitution in Bound-Propagation-Based Neural Network Verification

Panagiotis Kouvaros, Benedikt Brückner, Patrick Henriksen, Alessio Lomuscio
AAAI 2025
DOICopied
@inproceedings{kouvaros2025dynamic,
  author        = {Panagiotis Kouvaros and Benedikt Br{\"u}ckner and Patrick
                   Henriksen and Alessio Lomuscio},
  editor        = {Toby Walsh and Julie Shah and Zico Kolter},
  title         = {Dynamic Back-Substitution in Bound-Propagation-Based Neural
                   Network Verification},
  booktitle     = {Thirty-Ninth {AAAI} Conference on Artificial Intelligence,
                   Thirty-Seventh Conference on Innovative Applications of
                   Artificial Intelligence, Fifteenth Symposium on Educational
                   Advances in Artificial Intelligence, {AAAI} 2025,
                   Philadelphia, PA, USA, February 25 - March 4, 2025},
  pages         = {27383--27391},
  publisher     = {{AAAI} Press},
  year          = {2025},
  doi           = {10.1609/aaai.v39i26.34949},
  url           = {https://doi.org/10.1609/aaai.v39i26.34949},
}
2023

SHAPE: A Framework for Evaluating the Ethicality of Influence

Elfia Bezou-Vrakatseli, Benedikt Brückner, Luke Thorburn
EUMAS 2023
DOIarXivCopied
@inproceedings{bezouvrakatseli2023shape,
  author        = {Elfia Bezou-Vrakatseli and Benedikt Br{\"u}ckner and Luke
                   Thorburn},
  editor        = {Vadim Malvone and Aniello Murano},
  title         = {{SHAPE}: {A} Framework for Evaluating the Ethicality of
                   Influence},
  booktitle     = {Multi-Agent Systems - 20th European Conference, {EUMAS}
                   2023, Naples, Italy, September 14-15, 2023, Proceedings},
  series        = {Lecture Notes in Computer Science},
  volume        = {14282},
  pages         = {167--185},
  publisher     = {Springer},
  year          = {2023},
  doi           = {10.1007/978-3-031-43264-4\_11},
  url           = {https://doi.org/10.1007/978-3-031-43264-4\_11},
  eprint        = {2309.04352},
  archivePrefix = {arXiv},
}

A Semidefinite Relaxation Based Branch-and-Bound Method for Tight Neural Network Verification

Jianglin Lan, Benedikt Brückner, Alessio Lomuscio
AAAI 2023
DOICopied
@inproceedings{lan2023semidefinite,
  author        = {Jianglin Lan and Benedikt Br{\"u}ckner and Alessio Lomuscio},
  editor        = {Brian Williams and Yiling Chen and Jennifer Neville},
  title         = {A Semidefinite Relaxation Based Branch-and-Bound Method for
                   Tight Neural Network Verification},
  booktitle     = {Thirty-Seventh {AAAI} Conference on Artificial
                   Intelligence, {AAAI} 2023, Thirty-Fifth Conference on
                   Innovative Applications of Artificial Intelligence, {IAAI}
                   2023, Thirteenth Symposium on Educational Advances in
                   Artificial Intelligence, {EAAI} 2023, Washington, DC, USA,
                   February 7-14, 2023},
  pages         = {14946--14954},
  publisher     = {{AAAI} Press},
  year          = {2023},
  doi           = {10.1609/aaai.v37i12.26745},
  url           = {https://doi.org/10.1609/aaai.v37i12.26745},
}
2022

Siamese Basis Function Networks for Data-Efficient Defect Classification in Technical Domains

Tobias Schlagenhauf, Faruk Yildirim, Benedikt Brückner
AI4EA Workshop, SEFM 2022 Workshop
DOIarXivCopied
@inproceedings{schlagenhauf2022siamese,
  author        = {Tobias Schlagenhauf and Faruk Yildirim and Benedikt
                   Br{\"u}ckner},
  editor        = {Paolo Masci and Cinzia Bernardeschi and Pierluigi Graziani
                   and Mario Koddenbrock and Maurizio Palmieri},
  title         = {Siamese Basis Function Networks for Data-Efficient Defect
                   Classification in Technical Domains},
  booktitle     = {Software Engineering and Formal Methods. {SEFM} 2022
                   Collocated Workshops - AI4EA, F-IDE, CoSim-CPS, CIFMA,
                   Berlin, Germany, September 26-30, 2022, Revised Selected
                   Papers},
  series        = {Lecture Notes in Computer Science},
  volume        = {13765},
  pages         = {71--92},
  publisher     = {Springer},
  year          = {2022},
  doi           = {10.1007/978-3-031-26236-4\_7},
  url           = {https://doi.org/10.1007/978-3-031-26236-4\_7},
  eprint        = {2012.01338},
  archivePrefix = {arXiv},
}

Awards and service

2025
Top Reviewer Award
NeurIPS
2025
Lord Williams of Elvel Prize
Entente Cordiale Day Association
2023
Student Travel Scholarship
AAAI

Reviewer for NeurIPS (2024–2026), ICLR (2025–2026), ICML (2025), AAAI (2025–2026) and BMVC (2025).