What is the best coverage criteria?

What is the most effective mutation operators?

Regression testing has long been studied in “traditional” software systems. What does regression testing mean when I have a ML based system?

What is a regression bug?

How to apply testing/analysis in different applications domains? For example in DL based code analysis, security solutions (malware detection or network intrustion detections)

Techniques have been proposed to repair a deep neural network (e.g., by modifying the weights). In what scenarios or tasks do you think are such techniques more applicable?

How to achieve model interpretability? Examples like making surrogate models (e.g. in LIME), or a gradient based approach, as in Integrated Gradients, or a global approach such as rule induction.