Paul Temple and Jean-Marc Jézéquel were at DeepTest, perhaps the most attended workshop at ICSE. They have presented “Applying Multimorphic Testing to Deep Learning Systems”. It’s an ongoing work and the workshop was informal.
Multimorphic testing (MMT) aims to assess the effectiveness of a test suite for revealing performance variations of a system. We’ve promising empirical results in different contexts/domains see ICSE’18 poster and PhD thesis of Paul. The question is now: does MMT work for deep neural network-based systems?