Machine learning testing in an ADAS case study using simulation‐integrated bio‐inspired search‐based testing

This paper presents an extended version of Deeper, a search-based simulation-integrated test solution that generates failure-revealing scenarios for testing deep neural network-based lane-keeping system. In the newly proposed version, we utilize new set bio-inspired search algorithms, genetic algorithm (GA), ( μ + λ ) $$ \left(\mu +\lambda \right) and , \left(\mu, \lambda evolution strategies (ES), particle swarm optimization (PSO), leverage quality population seed domain-specific crossover mutation operations tailored presentation model used modeling scenarios. order to demonstrate capabilities generators within carry out empirical evaluation comparison with regard results five participating tools in cyber-physical systems competition at SBST 2021. Our shows Deeper not only represent considerable improvement on previous but also prove be effective efficient provoking number diverse ML-driven They can trigger several failures while promoting scenario diversity, under limited time budget, high target failure severity, strict speed limit constraints.