Fabio Arnez - Website

I’m a research engineer in the Laboratory of Embedded and Autonomous Systems (LSEA) at CEA-List, Université Paris-Saclay, France, where I lead research on trustworthy deep learning for automated vehicles and robots. I supervise PhD students and contribute to industry-driven AI projects funded by the European Union and the French government.

My research spans uncertainty quantification and runtime monitoring of neural networks for automated driving and aerial navigation, out-of-distribution and distribution-shift detection in vision components, and the dependability of automated systems. More recently, my research has extended to hallucination detection in large language models and to the reliability and robustness of quantized neural networks, including vision-language models. I also study world models, from their theoretical foundations to their robustness. I serve the research community as a reviewer for top-tier venues, including ICML, NeurIPS, ICLR, CVPR, AAAI, and UAI, and I was recognized as a Gold Reviewer at ICML 2026.

I hold a PhD in Computer Science from Université Paris-Saclay, with a thesis on uncertainty monitoring for safe automated navigation. My journey began with a BSc in electronics and telecommunications from Universidad Privada Boliviana (UPB), in my home country, Bolivia. I then won the RETECA foundation scholarship to pursue an MSc in embedded systems and microelectronics at the University of Applied Sciences and Arts of Southern Switzerland (SUPSI).

news

Sep 25, 2026	Our paper “Disentangling the Good From the Bad: Quantization-Induced Flips Are Not Random” was accepted at NeurIPS 2026 (December 2026, Sydney, Australia, with satellite events in Atlanta and Paris)!!! Disentangling the Good From the Bad: Quantization-Induced Flips Are Not Random Authors: Aymen Bouguerra, Alexandra Gomez-Villa, Chokri Mraidha, and Fabio Arnez NeurIPS 2026, Sydney, Atlanta & Paris.
Aug 31, 2026	Two of our papers were accepted at IEEE ICVES 2026 (November 9-11, 2026, Cochabamba, Bolivia)!!! Quantization and Corruption Robustness in Deployed Road-Scene Perception Authors: Aymen Bouguerra, Ansgar Radermacher, Chokri Mraidha, and Fabio Arnez Drive-SynOOD-OD: A Diffusion-Inpainted Synthetic Out-of-Distribution Object Detection Benchmark for Autonomous Driving Authors: Daniel Montoya, Mauricio Sayri Espinoza Mayzer, and Fabio Arnez IEEE ICVES 2026, Cochabamba, Bolivia.
Jul 15, 2026	Our new preprint “The SIGReg Objective as Variational Free Energy: A Theoretical Active-Inference Account of JEPA World Models” is now available on arXiv! The SIGReg Objective as Variational Free Energy: A Theoretical Active-Inference Account of JEPA World Models Authors: Fabio Arnez and Alexandra Gomez-Villa Preprint on arXiv.
May 13, 2026	Honored to be recognized as a Gold Reviewer at ICML 2026, placing among the top reviewers for this year’s conference!
Apr 30, 2026	Our paper “Less Precise Can Be More Reliable: A Systematic Evaluation of Quantization’s Impact on VLMs Beyond Accuracy” was accepted at ICML 2026 (July 6-11, 2026, Seoul, South Korea)!!! Less Precise Can Be More Reliable: A Systematic Evaluation of Quantization’s Impact on VLMs Beyond Accuracy Authors: Aymen Bouguerra, Daniel Montoya, Alexandra Gomez-Villa, Chokri Mraidha, and Fabio Arnez ICML 2026, Seoul, South Korea.

latest posts

Nov 03, 2025	Quantifying and Using Uncertainty in Deep Learning-based UAV Navigation
Apr 13, 2020	Probability & Statistics Fundamentals: Part I
Apr 07, 2020	Notes on Simple Approximate Inference in Bayesian Neural Networks

selected publications

Disentangling the Good From the Bad: Quantization-Induced Flips Are Not Random

Aymen Bouguerra, Alexandra Gomez-Villa, Chokri Mraidha, and Fabio Arnez

In Advances in Neural Information Processing Systems (NeurIPS), Dec 2026

To appear

Bib

@inproceedings{bouguerra2026disentangling,
  title = {Disentangling the Good From the Bad: Quantization-Induced Flips Are Not Random},
  author = {Bouguerra, Aymen and Gomez-Villa, Alexandra and Mraidha, Chokri and Arnez, Fabio},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year = {2026},
  month = dec,
  note = {To appear}
}

Less Precise Can Be More Reliable: A Systematic Evaluation of Quantization’s Impact on VLMs Beyond Accuracy

Aymen Bouguerra, Daniel Montoya, Alexandra Gomez-Villa, Chokri Mraidha, and Fabio Arnez

In International Conference on Machine Learning (ICML), Jul 2026

Bib HTML PDF Code Poster Slides

@inproceedings{bouguerra2025precisereliablesystematicevaluation,
  title = {Less Precise Can Be More Reliable: A Systematic Evaluation of Quantization's Impact on VLMs Beyond Accuracy},
  author = {Bouguerra, Aymen and Montoya, Daniel and Gomez-Villa, Alexandra and Mraidha, Chokri and Arnez, Fabio},
  booktitle = {International Conference on Machine Learning (ICML)},
  year = {2026},
  month = jul,
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://icml.cc/virtual/2026/poster/63466},
  pdf = {https://arxiv.org/pdf/2509.21173},
}

FindMeIfYouCan: Bringing Open Set metrics to near, far and farther Out-of-Distribution Object Detection

Daniel Montoya, Aymen Bouguerra, Alexandra Gomez-Villa, and Fabio Arnez

2025

Bib PDF

@misc{montoya2025findmeifyoucanbringingopenset,
  title = {FindMeIfYouCan: Bringing Open Set metrics to near, far and farther Out-of-Distribution Object Detection},
  author = {Montoya, Daniel and Bouguerra, Aymen and Gomez-Villa, Alexandra and Arnez, Fabio},
  year = {2025},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2506.14008},
  pdf = {https://arxiv.org/pdf/2506.14008}
}

Latent representation entropy density for distribution shift detection

Fabio Arnez, Daniel Alfonso Montoya Vasquez, Ansgar Radermacher, and François Terrier

In Conference on Uncertainty in Artificial Intelligence (UAI), 2024

Bib PDF

@inproceedings{arnez2024latent,
  title = {Latent representation entropy density for distribution shift detection},
  author = {Arnez, Fabio and Vasquez, Daniel Alfonso Montoya and Radermacher, Ansgar and Terrier, Fran{\c{c}}ois},
  booktitle = {Conference on Uncertainty in Artificial Intelligence (UAI)},
  year = {2024},
  url = {https://proceedings.mlr.press/v244/arnez24a.html},
  pdf = {https://raw.githubusercontent.com/mlresearch/v244/main/assets/arnez24a/arnez24a.pdf}
}

Deep neural network uncertainty runtime monitoring for robust and safe AI-based automated navigation

Fabio Alejandro Arnez Yagualca

Dec 2023

Bib PDF

@phdthesis{arnezyagualca:tel-04672736,
  title = {{Deep neural network uncertainty runtime monitoring for robust and safe AI-based automated navigation}},
  author = {Arnez Yagualca, Fabio Alejandro},
  url = {https://theses.hal.science/tel-04672736},
  number = {2023UPASG100},
  school = {{Universit{\'e} Paris-Saclay}},
  year = {2023},
  month = dec,
  keywords = {Deep neural network ; Uncertainty ; Monitoring ; Propagation ; Automated system ; R{\'e}seaux neuronaux profonds ; Incertitude ; Surveillance du temps d'ex{\'e}cution ; Propagation ; Syst{\`e}me automatis{\'e}},
  type = {Theses},
  pdf = {https://theses.hal.science/tel-04672736v1/file/122361_ARNEZ_YAGUALCA_2023_archivage.pdf},
  hal_id = {tel-04672736},
  hal_version = {v1}
}

Quantifying and using system uncertainty in uav navigation

Fabio Arnez, Ansgar Radermacher, and Huascar Espinoza

arXiv preprint arXiv:2206.01953, 2022

Bib

@article{arnez2022quantifying,
  title = {Quantifying and using system uncertainty in uav navigation},
  author = {Arnez, Fabio and Radermacher, Ansgar and Espinoza, Huascar},
  journal = {arXiv preprint arXiv:2206.01953},
  year = {2022}
}

Towards dependable autonomous systems based on bayesian deep learning components

Fabio Arnez, Huascar Espinoza, Ansgar Radermacher, and François Terrier

In 2022 18th European Dependable Computing Conference (EDCC), 2022

Bib

@inproceedings{arnez2022towards,
  title = {Towards dependable autonomous systems based on bayesian deep learning components},
  author = {Arnez, Fabio and Espinoza, Huascar and Radermacher, Ansgar and Terrier, Fran{\c{c}}ois},
  booktitle = {2022 18th European Dependable Computing Conference (EDCC)},
  pages = {65--72},
  year = {2022},
  organization = {IEEE}
}

news

Disentangling the Good From the Bad: Quantization-Induced Flips Are Not Random

Quantization and Corruption Robustness in Deployed Road-Scene Perception

Drive-SynOOD-OD: A Diffusion-Inpainted Synthetic Out-of-Distribution Object Detection Benchmark for Autonomous Driving

The SIGReg Objective as Variational Free Energy: A Theoretical Active-Inference Account of JEPA World Models

Less Precise Can Be More Reliable: A Systematic Evaluation of Quantization’s Impact on VLMs Beyond Accuracy

latest posts

selected publications