Deep Neural Networks (DNNs) are applied in several safety-critical domains and their trustworthiness is of paramount importance. For example, DNNs used in autonomous driving as classifiers should not misclassify detected objects; however, since obtaining perfect accuracy is not possible, special attention should be given to the most critical cases, e.g., pedestrians. This has been confirmed by the consortium of our partners from the automotive domain that provided us with specific risk levels for different misclassifications. A recent approach to improve DNN performance is to localise DNN weights responsible for the misclassifications and then adjust (repair) them to improve the misclassifications. However, they under-perform when they need to consider multiple misclassifications, and they do not consider the risk levels of the different misclassifications. To tackle this, we propose DistrRep, a distributed repair approach that first finds the best fixes for each critical misclassification, and then integrates them in a single repaired DNN model, by considering the risk levels. We assess DistrRep over three DNN models and a dataset of autonomous driving images, by considering requirements specified by our industrial partners. Experiments show that DistrRep is more effective than baseline approaches based on retraining, and other risk-unaware repair approaches.
Mon 17 AprDisplayed time zone: Dublin change
16:00 - 18:00 | Session 5: Testing AI/ML systemsResearch Papers / Previous Editions at Grand canal Chair(s): Jie M. Zhang King's College London | ||
16:00 20mTalk | Robustness assessment and improvement of a neural network for blood oxygen pressure estimation Previous Editions Paolo Arcaini National Institute of Informatics
, Andrea Bombarda University of Bergamo, Silvia Bonfanti University of Bergamo, Angelo Gargantini University of Bergamo, Daniele Gamba AISent S.r.l., Rita Pedercini AISent S.r.l. DOI | ||
16:20 20mTalk | An Empirical Evaluation of Mutation Operators for Deep Learning Systems Previous Editions DOI | ||
16:40 20mTalk | Distributed Repair of Deep Neural Networks Research Papers Davide Li Calsi Politecnico di Milano, Matias Duran National Institute of Informatics, Xiao-Yi Zhang School of Computer and Communication Engineering, University of Science and Technology Beijing, Paolo Arcaini National Institute of Informatics
, Fuyuki Ishikawa National Institute of Informatics | ||
17:00 20mTalk | Mutation Testing of Deep Reinforcement Learning Based on Real Faults Research Papers Florian Tambon Polytechnique Montréal, Vahid Majdinasab Polytechnique Montréal, Amin Nikanjam École Polytechnique de Montréal, Foutse Khomh Polytechnique Montréal, Giuliano Antoniol Polytechnique Montréal Pre-print | ||
17:20 20mTalk | Repairing DNN Architecture: Are We There Yet? Research Papers Jinhan Kim KAIST, Nargiz Humbatova USI Lugano, Gunel Jahangirova King's College London, Paolo Tonella USI Lugano, Shin Yoo KAIST Pre-print |