Michał Szczepankiewicz

Bartlomiej Olber, Krystian Radlak, Adam Popowicz et al.

Deep neural networks (DNN) have outstanding performance in various applications. Despite numerous efforts of the research community, out-of-distribution (OOD) samples remain a significant limitation of DNN classifiers. The ability to identify previously unseen inputs as novel is crucial in safety-critical applications such as self-driving cars, unmanned aerial vehicles, and robots. Existing approaches to detect OOD samples treat a DNN as a black box and evaluate the confidence score of the output predictions. Unfortunately, this method frequently fails, because DNNs are not trained to reduce their confidence for OOD inputs. In this work, we introduce a novel method for OOD detection. Our method is motivated by theoretical analysis of neuron activation patterns (NAP) in ReLU-based architectures. The proposed method does not introduce a high computational overhead due to the binary representation of the activation patterns extracted from convolutional layers. The extensive empirical evaluation proves its high performance on various DNN architectures and seven image datasets.

Krystian Radlak, Michał Szczepankiewicz, Tim Jones et al.

Machine learning (ML) algorithms generate a continuous stream of success stories from various domains and enable many novel applications in safety-critical systems. With the advent of autonomous driving, ML algorithms are being used in the automotive domain, where the applicable functional safety standard is ISO 26262. However, requirements and recommendations provided by ISO 26262 do not cover specific properties of machine learning algorithms. Therefore, specific aspects of ML (e.g., dataset requirements, performance evaluation metrics, lack of interpretability) must be addressed within some work products, which collect documentation resulting from one or more associated requirements and recommendations of ISO 26262. In this paper, we propose how key technical aspects and supporting processes related to development of ML-based systems can be organized according to ISO 26262 phases, sub-phases, and work products. We follow the same approach as in the ISO/PAS 21448 standard, which complements ISO 26262, in order to account for edge cases that can lead to hazards not directly caused by system failure.%, but resulting from functional insufficiencies of the intended functionality or by reasonably foreseeable misuse by persons.