[
https://issues.apache.org/jira/browse/SINGA-120?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15085612#comment-15085612
]
ASF subversion and git services commented on SINGA-120:
-------------------------------------------------------
Commit 9705f39cb4ed2247889bda760f2352b538321c35 in incubator-singa's branch
refs/heads/master from [~flytosky]
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=9705f39 ]
SINGA-120 - Implemented GRU and BPTT
Improve the code for computing \hat{h_t}, i.e., the new memory of gru unit.
The loss decreases under 2 per unit.
Update test files, including test for unrolling and gru unit (ComputeFeature
part)
> Implemented GRU and BPTT
> ------------------------
>
> Key: SINGA-120
> URL: https://issues.apache.org/jira/browse/SINGA-120
> Project: Singa
> Issue Type: New Feature
> Reporter: Ju Fan
>
> This ticket will implement the GRU model (one of the RNN models) and BPTT
> algorithm for training the GRU model.
> Particularly, it implements the GRU unit and unrolling the user defined
> model. Truncated BPTT is implemented as a subclass of Worker for computing
> the gradients of parameters.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
