Structuring Validation Targets of a Machine Learning Function Applied to Automated Driving
The validation of highly automated driving vehicles is an important challenge to the automotive industry, since even if the system is free from internal faults, its behaviour might still vary from the original intent. Reasons for these deviations from the intended functionality can be found in the unpredictability of environmental conditions as well the intrinsic uncertainties of the Machine Learning (ML) functions used to make sense of this complex input space.
In this paper, we propose a safety assurance case for a pedestrian detection function, a safety-relevant baseline functionality for an automated driving system. Our safety assurance case is presented in the graphical structuring notation (GSN) and combines our arguments against the problems of underspecification , the semantic gap , and the deductive gap .
KeywordsSafety Intended functionality Functional insufficiency Nominal performance Automated driving Machine learning Assurance case GSN
- 2.Amarnath, R., Munk, P., Thaden, E., Nordmann, A., Burton, S.: Dependability challenges in the model-driven engineering of automotive systems. In: 2016 IEEE International Symposium on Software Reliability Engineering Workshops, ISSREW, pp. 1–4, October 2016Google Scholar
- 3.Bergenhem, C., et al.: How to reach complete safety requirement refinement for autonomous vehicles. Technical report, CARS 2015 - Critical Automotive Applications: Robustness & Safety, Paris, France, September 2015Google Scholar
- 5.Evtimov, I., et al.: Robust physical-world attacks on deep learning models. Cryptography and Security (2017). https://arxiv.org/abs/1707.08945
- 6.Goodfellow, I.J., Shlens, J., Szegedy, C.: Explaining and harnessing adversarial examples (2015). https://arxiv.org/abs/1412.6572
- 7.Katz, G., Barrett, C., Dill, D., Julian, K., Kochenderfer, M.: Reluplex: an efficient SMT solver for verifying deep neural networks. Technical report. Stanford University, USA (2017). https://arxiv.org/pdf/1702.01135.pdf
- 8.Kendall, A., Gal, Y.: What uncertainties do we need in Bayesian deep learning for computer vision? (2017). http://arxiv.org/abs/1703.04977
- 10.Krizhevsky, A., Sutskever, I., Hinton., G.E.: ImageNet classification with deep convolutional neural networks. In: Advances in Neural Information Processing Systems, pp. 1097–1105 (2012)Google Scholar
- 12.Leveson, N.G.: Engineering a Safer World: Systems Thinking Applied to Safety. The MIT Press, Cambridge (2011)Google Scholar
- 13.Nguyen, A.M., Yosinski, J., Clune, J.: Multifaceted feature visualization: uncovering the different types of features learned by each neuron in deep neural networks. CoRR abs/1602.03616 (2016). http://arxiv.org/abs/1602.03616
- 14.Tas, Ö.Ş., Kuhnt, F., Zllner, J.M., Stiller, C.: Functional system architectures towards fully automated driving. In: 2016 IEEE Intelligent Vehicles Symposium, IV (2016). http://ieeexplore.ieee.org/document/7535402/