The increasing use of deep neural networks (DNNs) in safety-critical systems
has raised concerns about their potential for exhibiting ill-behaviors. While
DNN verification and testing provide post hoc conclusions regarding unexpected
behaviors, they do not prevent the erroneous behavio