@inproceedings{magnussen2024adaptive,
  abstract = {Almost all sensors suffer from some level of uncertainty introduced from production inaccuracies. When the sensor data is processed by machine learning, quantifying the impact of such production inaccuracies on the output of the machine learning model becomes difficult.

Certain neural network architectures, such as continuous feature networks, allow individual features and data to be omitted while still being able to correctly predict the result without the need for retraining. Such features can, for example, be individual channels of a sensor. This article proposes a method to use the capability to omit arbitrary features or sensor channels to calculate Shapley values for each sensor channel. These Shapley values represent the contribution of each individual channel to the measurement. They are defined using an arbitrary function called the ``value function''. If the value function is defined as the error of the current measurement, the Shapley values will represent the contribution of each sensor channel to the error of the measurement result.

By calculating Shapley values like this for a large unlabelled dataset of measurements, it is possible to understand how much measurement error was introduced by which channel of which sensor in each measurement. Averaging the Shapley values for each sensor in the dataset will then result in a metric for each channel of that sensor, which represents a contribution to measurement errors. By comparing these values to any arbitrary quality metrics for the sensor channels obtained in a calibration process or similar step, it is possible to correlate and quantify which value in the quality metric will cause how much of a measurement error, or whether the quality metric is even relevant for the measurement accuracy.

This article will show the efficacy and use case of the method on an example of the production and quality control of optical sensors based on multiple spatially resolved reflection spectroscopy.},
  author = {Magnussen, Birk Martin and Jessulat, Maik and Stern, Claudius and Sick, Bernhard},
  booktitle = {International Conference on Big Data Analytics (ICBDA)},
  interhash = {f09ad8129c185efdb4db7a29e7d44939},
  intrahash = {1a6671c789facc1cdcc1e4d3ea4b1d4d},
  note = {(accepted)},
  publisher = {IEEE},
  title = {Adaptive Shapley: Using Explainable AI with Large Datasets to Quantify the Impact of Arbitrary Error Sources},
  year = 2024
}

@inproceedings{schreck2023height,
  abstract = {We present a novel method for free space detection in dynamic environments like factory sites, crucial for autonomous forklifts to avoid collisions. It introduces a technique for fast surface normal estimation using spherical projected LiDAR data, which helps in detecting free space efficiently and in real-time. The method's effectiveness is proven with a 50.90% mIoU score on the Semantic KITTI dataset at 105 Hz, and a 63.30% mIoU score on a factory site dataset at 54 Hz.},
  author = {Schreck, Steven and Reichert, Hannes and Hetzel, Manuel and Doll, Konrad and Sick, Bernhard},
  booktitle = {International Conference on Control, Mechatronics and Automation (ICCMA)},
  interhash = {d3c0bbf124344c070311798efc0d96cd},
  intrahash = {c68e47214fba4ebbe449b410278ca949},
  note = {(accepted)},
  title = {Height Change Feature Based Free Space Detection},
  year = 2023
}

@inproceedings{huang2023active,
  abstract = {Active learning strategies aim to minimize the number of queried samples for model training. However, two challenges in pool-based deep active learning on imbalanced datasets are observed in experiments: (1) the declining performance of active learning strategies due to imbalanced class distribution; (2)~the lack of sample diversity in acquisition batches due to the absence of timely model updates. This paper proposes the AL-FaMoUS, a general solution combining fast model updates and class-balanced minibatch selection to the active learning process. Furthermore, a simplification of the AL-FaMoUS, which selects one single sample in each acquisition minibatch, is experimentally evaluated on four image and three time-series imbalanced datasets. The results demonstrate that the implemented AL-FaMoUS outperforms the other adopted AL strategies, including uncertainty sampling and BALD solely combined with either the fast model update or the class balance selection strategy, in terms of accuracy and Macro F1 score.},
  author = {Huang, Zhixin and He, Yujiang and Herde, Marek and Huseljic, Denis and Sick, Bernhard},
  booktitle = {Workshop on Interactive Adapative Learning (IAL), ECML PKDD},
  interhash = {628e2871115e97a27bb1eee0484e1209},
  intrahash = {53a9d44816a8f463ce66e8efa287fb0b},
  pages = {28--45},
  title = {Active Learning with Fast Model Updates and Class-Balanced Selection for Imbalanced Datasets},
  url = {https://ceur-ws.org/Vol-3470/paper5.pdf},
  year = 2023
}

@article{heidecker2023corner,
  abstract = {Applications using machine learning (ML), such as highly autonomous driving, depend highly on the performance of the ML model. The data amount and quality used for model training and validation are crucial. If the model cannot detect and interpret a new, rare, or perhaps dangerous situation, often referred to as a corner case, we will likely blame the data for not being good enough or too small in number. However, the implemented ML model and its associated architecture also influence the behavior. Therefore, the occurrence of prediction errors resulting from the ML model itself is not surprising. This work addresses a corner case definition from an ML model's perspective to determine which aspects must be considered. To achieve this goal, we present an overview of properties for corner cases that are beneficial for the description, explanation, reproduction, or synthetic generation of corner cases. To define ML corner cases, we review different considerations in the literature and summarize them in a general description and mathematical formulation, whereby the expected relevance-weighted loss is the key to distinguishing corner cases from common data. Moreover, we show how to operationalize the corner case characteristics to determine the value of a corner case. To conclude, we present the extended taxonomy for ML corner cases by adding the input, model, and deployment levels, considering the influence of the corner case properties.},
  author = {Heidecker, Florian and {Bieshaar, Maarten und Sick}, Bernhard},
  interhash = {3258033e63148a1e9aa4d4e502db4f1b},
  intrahash = {e2e0ed2d4af801ab96e8920becbbb9d4},
  journal = {AI Perspectives & Advances},
  note = {(accepted)},
  title = {Corner Cases in Machine Learning Processes},
  year = 2023
}

@inproceedings{breitenstein2023what,
  abstract = {In safety-critical applications such as automated driving, perception errors may create an imminent risk to vulnerable road users (VRU). To mitigate the occurrence of unexpected and potentially dangerous situations, so-called corner cases, perception models are trained on a huge amount of data. However, the models are typically evaluated using task-agnostic metrics, which do not reflect the severity of safety-critical misdetections. Consequently, misdetections with particular relevance for the safe driving task should entail a more severe penalty during evaluation to pinpoint corner cases in large-scale datasets. In this work, we propose a novel metric IoUw that exploits relevance on the pixel level of the semantic segmentation output to extend the notion of the intersection over union (IoU) by emphasizing small areas of an image affected by corner cases. We (i) employ IoUw to measure the effect of pre-defined relevance criteria on the segmentation evaluation, and (ii) use the relevance-adapted IoUw to refine the identification of corner cases. In our experiments, we investigate vision-based relevance criteria and physical attributes as per-pixel criticality to factor in the imminent risk, showing that IoUw precisely accentuates the relevance of corner cases.},
  author = {Breitenstein, Jasmin and Heidecker, Florian and Lyssenko, Maria and Bogdoll, Daniel and Bieshaar, Maarten and Zöllner, J. Marius and Sick, Bernhard and Fingscheidt, Tim},
  booktitle = {Workshop on roBustness and Reliability of Autonomous Vehicles in the Open-world (BRAVO), ICCV},
  interhash = {82fd272efc4669189c8e047fa90b6db9},
  intrahash = {dd6f5a976f1b9423ffe3a36248745b42},
  note = {(accepted)},
  pages = {3991--4000},
  title = {What Does Really Count? Estimating Relevance of Corner Cases for Semantic Segmentation in Automated Driving},
  url = {https://openaccess.thecvf.com/content/ICCV2023W/BRAVO/papers/Breitenstein_What_Does_Really_Count_Estimating_Relevance_of_Corner_Cases_for_ICCVW_2023_paper.pdf},
  year = 2023
}