3
$\begingroup$

When training a relatively small DL model, which takes several hours to train, I typically start with some starting points from literature and then use a trial-and-error or grid-search approach to fine-tune the values of the hyper-parameters, in order to prevent overfitting and achieve sufficient performance.

However, it is not uncommon for large models to have training time measured in days or weeks [1], [2], [3].

How are hyperparameters determined in such cases?

$\endgroup$
1
  • 1
    $\begingroup$ I've seen some papers which are deriving a small dataset which has more or less the same characteristics as the large one and do the hyper-parameter selection on it. $\endgroup$ – razvanc92 Jul 24 '20 at 14:10
3
$\begingroup$

In general, it is definitely very computationally expensive, so an exhaustive search is not performed in practice. However, there are some recent approaches for determining whether the architecture is "fine" without training the neural network first - by looking at the covariance matrix after forwarding the data, for example, in a recent paper Neural Architecture Search without Training. However, such an approach is very limited.

$\endgroup$

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.