This may sound silly: will the CNN overtrain? From what I have seen so far, having many epochs and parameters are the key of success of CNN. And there is almost nothing I found is about overtrain a CNN.

However, I have encountered some issue with my dataset. I keep training for a long epochs and the training accuracy is increasing. However, my validation set accuracy is large at epoch = 1 and then start to decrease (i ran for about 30 epochs).

Is this the problem of my dataset? Or I did something wrong? Or I overtrain the CNN? So less than 1 epoch is enough?

  • $\begingroup$ How pronounced is the decrease? Does it roughly fluctuate around the same accuracy or does the training accuracy actually decrease significantly? $\endgroup$
    – tsiki
    Nov 25, 2015 at 12:57
  • $\begingroup$ No, it , e.g, drops from 90% accuracy to 70% $\endgroup$ Nov 25, 2015 at 18:57

1 Answer 1


CNN, like any other neural network, overfits to the training data if it is trained for too long on the same training dataset. The purpose of the validation set is to stop training when performance on validation set starts decreasing, indicating that the model is overfitting the training data. Check this for more info.

  • $\begingroup$ Thanks for your answer. Yea, I refine my question and make it a bit clearer. $\endgroup$ Dec 4, 2015 at 19:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.