[jira] Issue Comment Edited: (MAHOUT-214) Implement Stacked RBM

["Olivier Grisel (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MAHOUT-214?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12800826#action_12800826 ] 

Olivier Grisel edited comment on MAHOUT-214 at 1/15/10 6:45 PM:
----------------------------------------------------------------

Stacked Autoencoders can achieve the same kind of semantic hashing purpose. Maybe with less computation:

 http://velblod.videolectures.net/2008/pascal2/icml08_helsinki/szummer_sslcdr/icml08_szummer_sslcdr_01.pdf
 (slides)
 http://www.cs.nyu.edu/~ranzato/publications/ranzato-icml08.pdf (paper)

However I don't see any trivial way to make them trainable as efficiently distributed  MapReduce jobs. Still worth trying.

      was (Author: ogrisel):
    Stacked Autoencoders can achieve the same kind of semantic hashing purpose. Maybe with less computation:

 http://velblod.videolectures.net/2008/pascal2/.../icml08_szummer_sslcdr_01.pdf (slides)
 http://www.cs.nyu.edu/~ranzato/publications/ranzato-icml08.pdf (paper)

However I don't see any trivial way to make them trainable as efficiently distributed  MapReduce jobs. Still worth trying.
  
> Implement Stacked RBM 
> ----------------------
>
>                 Key: MAHOUT-214
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-214
>             Project: Mahout
>          Issue Type: New Feature
>          Components: Clustering
>    Affects Versions: 0.2
>            Reporter: Prasen Mukherjee
>            Priority: Minor
>             Fix For: 0.3
>
>
> Implement Stacked RBM based models for generic data.  Suggested paper is :  http://www.cs.toronto.edu/~rsalakhu/papers/semantic_final.pdf For a more indepth understanding of  Stacked RBMs refer to : http://www.cs.toronto.edu/~hinton/papers.html
> http://jarbm.sourceforge.net/ could be a good starting point. 

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.