There is a Dataset of 65 tuples. I want to Augment new Data from this set and validate my ANN on the original Data. Is there a possibility, that my ANN already overfits on the augmentet Data. For example that the augmentation Is simulated by the ANN, and how does one prevent this ?
$\begingroup$
$\endgroup$
3
-
$\begingroup$ In theory it's possible, everything depends on your data, the model you're training and how you're going to augment your training data. $\endgroup$– Mikhail BerlinkovCommented Nov 13, 2018 at 15:34
-
$\begingroup$ you were right. it learned the augmentation instead of the Data. I solved the problem by splitting the Dataset before augmentation and augment only a test set, and validate on data , never seen by the system. $\endgroup$– Alexander VocaetCommented Nov 15, 2018 at 10:08
-
$\begingroup$ I'll post it as the answer then if you don't mind. $\endgroup$– Mikhail BerlinkovCommented Nov 16, 2018 at 14:23
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
In theory it's possible, everything depends on your data, the model you're training and how you're going to augment your training data.
