0
$\begingroup$

I'm currently using an autoencoder CNN that's built upon the VGG-16 architecture that was designed by someone else. I want to replicate their results using their dataset first but I'm finding that:

-Validation losses diverge from training losses fairly early on (I get to around 10 epochs and it already looks like it's overfitting) -At its best, the validation losses aren't even close to being as low as training losses -In general, the accuracy is still worse than reported in their paper.

I'm new to machine learning and want to know if there are hyperparameters I should try to change or what I can do to maybe tinker with it without changing its architecture?

$\endgroup$
0
$\begingroup$

Are you in fact using the same architecture as they are? If not that could potentially be the problem.

Otherwise, are you using the same trainings protocol as they, i.e. optimizer, learning rate, learning rate schedule, batch size, preprocessing, weight initialization, number of training epochs? Depending on the size of your model and the amount of training data, 10 epochs might not be enough to judge about your models performance.

Can you link the paper?

| improve this answer | |
$\endgroup$
  • $\begingroup$ Skimmed through the paper. Unfortunately there is no statement about how long they train the model, however form the code it seems like the default epoch parameter is 200. Indeed it could be the case the mobile data set used a different protocol. $\endgroup$ – Tinu May 14 at 5:21
  • $\begingroup$ You mentioned accuracy in your question. I assume you mean mean-squared-error? I don't see any accuracy metric in the paper. $\endgroup$ – Tinu May 14 at 5:22
  • $\begingroup$ Yes, I meant MSE. I think they also mentioned accuracy in the other dataset but not again for the mobile data. Would you have any ideas where to start with any kind of parameter modification? $\endgroup$ – nmtp May 14 at 15:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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