2 ms·

I would have hoped that ISO 26262 would help to prevent this sort of thing from happening. Maybe it is an audit quality issue?

• pmarreck 8y ago
  There is also likely difficulty in testing an ML black-box. When I took the ML class (Stanford's, online), they actually recommended setting aside some percentage of the training data as test data, but this is (IMHO) sub-par because it doesn't eliminate the possibility of things like overfitting