singa-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF subversion and git services (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SINGA-9) Add Support for Restricted Boltzman Machine (RBM) model
Date Fri, 28 Aug 2015 09:29:45 GMT

    [ https://issues.apache.org/jira/browse/SINGA-9?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14718264#comment-14718264
] 

ASF subversion and git services commented on SINGA-9:
-----------------------------------------------------

Commit ef4de796303550b1f3e31fd2fddd9eb831db2b06 in incubator-singa's branch refs/heads/master
from [~zhaojing]
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=ef4de79 ]

SINGA-9 Add Support for Restricted Boltzman Machine (RBM) model
This is to implement RBM in SINGA.
To training RBM models, the Contrastive Divergence (CD) algorithm is implemented.
We have implemented a BPWorker to run the Back-Propagation algorithm. To implement the CD
algorithm, we follow the same way
to create a CDWorker whose RunOneBatch function controls the logic of the CD algorithm, including
positive phase,
negative phase and computing gradient phase. RBM's layers are different to the layers for
feed-forward neural networks,
hence new layers for RBM models are added.


> Add Support for Restricted Boltzman Machine (RBM) model
> -------------------------------------------------------
>
>                 Key: SINGA-9
>                 URL: https://issues.apache.org/jira/browse/SINGA-9
>             Project: Singa
>          Issue Type: New Feature
>            Reporter: wangwei
>            Assignee: ZHAOJING
>
> This ticket is to implement RBM in SINGA.
> To training RBM models, the Contrastive Divergence (CD) algorithm should be implemented.

> We have implemented a BPWorker to run the Back-Propagation algorithm. To implement the
CD algorithm, we follow the same way to create a CDWorker whose RunOneBatch function controls
the logic of the CD algorithm, including positive phase, negative phase and computing gradient
phase.
> RBM's layers are different to the layers for feed-forward neural networks, hence new
layers for RBM models should be added.



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

Mime
View raw message