Robustness

A neural network is robust if it returns correct output on all input.

Impractical: the input space is too large to be coveredo

A learning model is locally-robust if it returns the
correct output on inputs similar to inputs in the training set.