Jens Brandt, PhD student at the THK-AI Research Cluster (https://thk-ai.de) at Technische Hochschule Köln, has published a research paper that will be presented at NeurIPS 2025 (https://neurips.cc), one of the most prestigious conferences in the field of artificial intelligence and machine learning. The conference takes place in December 2025 in San Diego.
The paper, titled „From Faults to Features: Pretraining to Learn Robust Representations against Sensor Failures,“ addresses a critical challenge in safety-critical applications such as autonomous vehicles. The research demonstrates how machine learning models can be made more robust against sensor failures through innovative pretraining techniques.
indicating driving direction). Middle: perturbed (˜x) vs. true (x) wheelspeed measurements.
Bottom: sideslip estimate (ˆy) under corruptions vs. true VSA (y).
This work is the result of a collaboration between Technische Hochschule Köln, The Leiden Institute of Advanced Computer Science (LIACS) at Leiden University (NL), Toyota Gazoo Racing, and the Toyota Research Institute. Jens Brandt’s PhD thesis is supervised by Prof. Dr. Thomas Bartz-Beielstein (THK-AI Research Cluster), Prof. Dr. Thomas Bäck (Leiden University, Netherlands), and Dr. Marc Hilbert (Toyota).
The paper and accompanying poster are available on the official NeurIPS website: https://neurips.cc/virtual/2025/loc/san-diego/poster/119529
About the Research
Machine learning models play a key role in safety-critical applications, where their robustness during inference is essential to ensure reliable operation. The research proposes a self-supervised masking scheme that simulates common sensor failures and explicitly trains models to recover the original signal. As a practical application, the method was deployed on a modified Lexus LC 500, demonstrating that the pretrained model successfully operates as a substitute for a physical sensor in a closed-loop control system for autonomous racing.
About THK-AI Research Cluster
The THK-AI Research Cluster at Technische Hochschule Köln (https://thk-ai.de) focuses on advancing artificial intelligence research and its practical applications, with particular emphasis on developing safe and reliable AI systems.
BibTeX
@inproceedings{bran25a,
author = {Brandt, Jens U. and P{\"u}tz, Noah C. and Greiff, Marcus and Lew, Thomas Jonathan and Subosits, John and Hilbert, Marc and Bartz-Beielstein, Thomas},
booktitle = {Advances in Neural Information Processing Systems},
date-modified = {2025-11-13 18:21:11 +0100},
note = {NeurIPS 2025 poster},
title = {From Faults to Features: Pretraining to Learn Robust Representations against Sensor Failures},
url = {https://openreview.net/pdf/f4af094c04a58154e07f3b45207fa5fc4a27b0e6.pdf},
year = {2025},
}