#global-explanation

Question 1: Single choice

A bank uses a neural network to decide on loans and must tell each rejected applicant why their own application was refused. What kind of explanation does this call for?

A local explanation of that applicant's individual prediction1
A global explanation of how the model behaves on average over all applicants2
A higher test accuracy reported on a held-out set of past applicants3
A sanity check showing that the network's saliency maps depend on its weights4

Timed test