Search

Fail-operational systems are a prerequisite for autonomous driving. Without a driver who can act as a fallback solution in a critical failure scenario, the system has to be able to mitigate failures on its own and keep critical applications operational. To reduce redundancy cost, graceful degradation can be applied by repurposing hardware resources at run-time. Critical applications can be kept operational by starting passive backups and shutting down non-critical applications instead to make sufficient resources available. In order to design such systems efficiently, the degradation effects on reliability and cost savings have to be analyzed. In this paper we present our approach to formally analyze the impact of graceful degradation on the reliability of critical and non-critical applications. We then quantify the effect of graceful degradation on the reliability of both critical and non-critical applications in distributed automotive systems and compare the achieved cost reduction with conventional redundancy approaches. In our experiments redundancy overhead could be reduced by 80% compared to active redundancy in a scenario with a balanced mix of critical and non-critical applications using our graceful degradation approach Overall, we present a detailed reliability and cost analysis of graceful degradation in distributed automotive systems. Our findings confirm that using graceful degradation can tremendously reduce cost compared to conventional redundancy approaches with no negative impact on the redundancy of critical applications if a reliability reduction of non-critical applications can be accepted. Our results show that a trade-off between the impact of the degradation on the reliability of non-critical applications and cost reduction has to be made.

A variety of methods exist to explain image classification models. However, whether they provide any benefit to users over simply comparing various inputs and the model's respective predictions remains unclear. We conducted a user study (N=240) to test how such a baseline explanation technique performs against concept-based and counterfactual explanations. To this end, we contribute a synthetic dataset generator capable of biasing individual attributes and quantifying their relevance to the model. In a study, we assess if participants can identify the relevant set of attributes compared to the ground-truth. Our results show that the baseline outperformed concept-based explanations. Counterfactual explanations from an invertible neural network performed similarly as the baseline. Still, they allowed users to identify some attributes more accurately. Our results highlight the importance of measuring how well users can reason about biases of a model, rather than solely relying on technical evaluations or proxy tasks. We open-source our study and dataset so it can serve as a blue-print for future studies. For code see, https://github.com/berleon/do_users_benefit_from_interpretable_vision, Published at ICLR 2022

A growing number of approaches exist to generate explanations for image classification. However, few of these approaches are subjected to human-subject evaluations, partly because it is challenging to design controlled experiments with natural image datasets, as they leave essential factors out of the researcher's control. With our approach, researchers can describe their desired dataset with only a few parameters. Based on these, our library generates synthetic image data of two 3D abstract animals. The resulting data is suitable for algorithmic as well as human-subject evaluations. Our user study results demonstrate that our method can create biases predictive enough for a classifier and subtle enough to be noticeable only to every second participant inspecting the data visually. Our approach significantly lowers the barrier for conducting human subject evaluations, thereby facilitating more rigorous investigations into interpretable machine learning. For our library and datasets see, https://github.com/mschuessler/two4two/, Comment: 6 pages, 3 figures, presented at the ICLR 2021 RAI workshop