A Algorithmic details A.1 Problem formulation for model repairment

This section extends the discussion on the problem formulation of model repairment. There are further considerations for executing and evaluating model repairment in practice. Generalisation and efficiency of model repairment In practice the number of failure cases might be large or even infinite. Gaussian noise may also fail on all the other inputs with such level of noise. One type of mistake might incur more costs than others (e.g., in medical applications, false Instead we present a "predictive" approach that removes this computational burden.