How would you model threats for a machine learning system at Asapp, including data poisoning and model inversion?
