2 ms·
Training on synthetic data has the problem that the models may drift away from the original distribution because errors could get amplified.
by cubefox 4y ago
Training on synthetic data has the problem that the models may drift away from the original distribution because errors could get amplified.