-
The need for systematic approaches in risk assessment of safety-critical AI-applications in machinery
Franziska Wolny, Silvia Vock, Rasmus Adler, Taras Holayad
Prceedings of the European Safey and Reliability Conference (ESREL), Stavanger/Norway,
June 15, 2025
The integration of artificial intelligence (AI) into safety-critical machinery applications in industrial environments presents substantial challenges for conformity assessment and safety certification.
Read more
-
Dynamic Risk Assessment for Human-Robot Collaboration Using a Heuristics-based Approach
Georgios Katranis, Frederik Plahl, Joachim Grimstadt, Ilshat Mamaev, Silvia Vock, Andrey Morozov
,
January 21, 2025
Human-robot collaboration (HRC) introduces significant safety challenges, particularly in protecting human operators working alongside collaborative robots (cobots). While current ISO standards emphasize risk assessment and hazard identification, these procedures are often insufficient for addressing the complexity of HRC environments, which involve numerous design factors and dynamic interactions.
Read more
-
From Classical to Advanced Risk Methods: Demonstrator for Industrial Cyber-Physical Systems
Andrey Morozov, Tagir Fabarisov, Silvia Vock, Thorben Schey, Artur Karimov, Georg Siedel, Joachim Grimstad, Arne Sonnenburg, Thomas Mossner
,
January 14, 2025
Modern industrial Cyber-Physical Systems (CPS) exhibit high levels of reconfigurability and heterogeneity, posing significant challenges for risk assessment in dynamic environments. Traditional risk assessment methods, originally developed for simpler systems
Read more
-
A Time-series Data Generation Tool for Risk Assessment of Robotic Applications
Yuliang Ma, Apurv Patel, Don Kurian, Julien Siebert, Silvia Vock, Andrey Morozov
,
January 12, 2025
Robotic systems increasingly rely on artificial intelligence (AI) to enhance their capabilities in performing complex tasks across various domains. The development and evaluation of AI systems usually require high-quality datasets
Read more
-
A practical approach to evaluating the adversarial distance for machine learning classifiers
Siedel, G., Gupta, E., Morozov, A.
Proceedings of the ASME 2024 International Mechanical Engineering Congress and Exposition (IMECE),
November 17, 2024
Adversarial distance is a highly informative measure of robustness that enjoys little attention in the adversarial robustness domain. We built a practical attack algorithm to evaluate this measure more effectively than implementations in the popular Adversarial Robustness Toolbox (ART). We also combine this upper-bound estimator with the popular lower-bound estimation method CLEVER, showing CLEVERs inability to calculate a correct lower bound in many cases.
Read more
-
Improved Dimensionality Dependence for Zeroth-Order Optimisation over Cross-Polytopes
Weijia Shao
,
July 29, 2024
This work proposes an algorithm improving the dimensionality dependence for gradient-free optimisation over cross-polytopes, which has many applications such as adversarial attacks, explainable AI and sparse regression.
Read more
-
Investigating the Corruption Robustness of Image Classifiers with Random p-norm Corruptions
Siedel, G., Shao, W., Vock, S., Morozov, A.
Proceedings of the 19th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications - Volume 2: VISAPP,
February 29, 2024
We show that image classifiers are not only vulnerable to invisible adversarial noise, but also to invisible random noise, and propose a corresponding metric. We also find that training data augmentation with a combination of p-norm noise types contributes more to corruption robustness than common noise injection.
Read more
-
Adaptive Zeroth-Order Optimisation of Nonconvex Composite Objectives
Weijia Shao, Sahin Albayrak
,
September 19, 2022
In this paper, we propose and analyse algorithms for zeroth-order optimisation of non-convex composite objectives, focusing on reducing the complexity dependence on dimensionality. This is achieved by exploiting the low dimensional structure of the decision set using the stochastic mirror descent method with an entropy alike function, which performs gradient descent in the space equipped with the maximum norm
Read more
-
Optimistic optimisation of composite objective with exponentiated update
Weijia Shao, Fikret Sivrikaya and Sahin Albayrak
,
August 22, 2022
This paper proposes a new family of algorithms for the online optimisation of composite objectives. The algorithms can be interpreted as the combination of the exponentiated gradient and p-norm algorithm.
Read more
-
Utilizing Class Separation Distance for the Evaluation of Corruption Robustness of Machine Learning Classifiers
Georg Siedel, Silvia Vock, Andrey Morozov, Stefan Voß
,
June 27, 2022
Robustness is a fundamental pillar of Machine Learning (ML) classifiers, substantially determining their reliability. Methods for assessing classifier robustness are therefore essential. In this work, we address the challenge of evaluating corruption robustness in a way that allows comparability and interpretability on a given dataset. We propose a test data augmentation method that uses a robustness distance derived from the datasets minimal class separation distance.
Read more