Hyperparameters for on-line learning

Supposing you have a neural network model who’s dataset is an infinite stream of inputs and targets from users surfing the Web, how do you choose its hyperparameters?


1 Response to “Hyperparameters for on-line learning”

  1. 1 Xavier Bouthillier February 4, 2013 at 12:31

    One can run the same model with different set of hyperparameters simultaneously on the same stream of inputs. During runtime, it is possible to filter out those with the worst error rate or generalization error rate. A lot of bad hyperparameters configurations can be dropped early on to avoid requiring too much computation power.

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: