You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@singa.apache.org by "wangwei (JIRA)" <ji...@apache.org> on 2016/05/26 10:12:12 UTC

[jira] [Created] (SINGA-183) Add the base classes for optimizer, constraint and regularizer

wangwei created SINGA-183:
-----------------------------

             Summary: Add the base classes for optimizer, constraint and regularizer
                 Key: SINGA-183
                 URL: https://issues.apache.org/jira/browse/SINGA-183
             Project: Singa
          Issue Type: New Feature
            Reporter: wangwei


Deep learning models are training via SGD algorithms to optimize the objective loss functions.
This ticket creates the base classes for the SGD algorithms, called Optimizer.
The Constraint class is created, which adds constraints on the gradients of parameters, e.g., the L2 norm should be within a threshold.
The Regularizer class is created, which implements L2 norm regularization for the objective loss function.

A simple vanilla SGD optimizer would be implemented for testing.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)