New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Explore different neural network models #78

Closed
bamos opened this Issue Jan 12, 2016 · 5 comments

Comments

Projects
None yet
4 participants
@bamos
Collaborator

bamos commented Jan 12, 2016

The original nn4 model was trained with a lot of data that we currently don't have,
so naturally a different neural network architecture is probably best-suited
for the data we're currently training on.
This idea is partially validated by the manually-written nn4.small models
performing better than the nn4 model.
This issue explores automatic ways of creating smaller networks
by using random hyper-parameter choices.

@bamos bamos changed the title from Explore different model sizes to Explore different neural network models Jan 12, 2016

@jmorra

This comment has been minimized.

Show comment
Hide comment
@jmorra

jmorra Jan 19, 2016

As a suggestion have the authors of this project considered something along the lines of this for hyper-parameter optimization?

jmorra commented Jan 19, 2016

As a suggestion have the authors of this project considered something along the lines of this for hyper-parameter optimization?

@ogrisel

This comment has been minimized.

Show comment
Hide comment
@ogrisel

ogrisel Jan 19, 2016

You might also be interested in a hosted service version of hyperopt with both Python and Lua client SDKs:

http://oscar.sensout.com/

ogrisel commented Jan 19, 2016

You might also be interested in a hosted service version of hyperopt with both Python and Lua client SDKs:

http://oscar.sensout.com/

@bamos bamos removed the help wanted label Mar 8, 2016

@Martyn10

This comment has been minimized.

Show comment
Hide comment
@Martyn10

Martyn10 Aug 5, 2016

The "Center Loss" function proposed here looks interesting. Seems like it would be easier to train compared to triplet loss:
http://ydwen.github.io/papers/WenECCV16.pdf

On his site he said he plans to release code in the future:
http://ydwen.github.io/

Martyn10 commented Aug 5, 2016

The "Center Loss" function proposed here looks interesting. Seems like it would be easier to train compared to triplet loss:
http://ydwen.github.io/papers/WenECCV16.pdf

On his site he said he plans to release code in the future:
http://ydwen.github.io/

@bamos

This comment has been minimized.

Show comment
Hide comment
@bamos

bamos Aug 5, 2016

Collaborator

Thanks for the reference @Martyn10! We'll further look into using the center loss.

Collaborator

bamos commented Aug 5, 2016

Thanks for the reference @Martyn10! We'll further look into using the center loss.

@stale

This comment has been minimized.

Show comment
Hide comment
@stale

stale bot Nov 18, 2017

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

stale bot commented Nov 18, 2017

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

@stale stale bot added the stale label Nov 18, 2017

@stale stale bot closed this Nov 25, 2017

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment