3 ms·
We know the tests are overfitting. That's not really a problem we can worry about too much; the tests can't be perfectly comprehensive and also finite. The prob
by naniwaduni 7y ago
We know the tests are overfitting. That's not really a problem we can worry about too much; the tests can't be perfectly comprehensive and also finite. The problem is that we're using to use the model to tell us something about adversarial input.