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 2018/03/05 05:58:00 UTC

[jira] [Created] (SINGA-343) Add neural net operations to Tensor

wangwei created SINGA-343:
-----------------------------

             Summary: Add neural net operations to Tensor
                 Key: SINGA-343
                 URL: https://issues.apache.org/jira/browse/SINGA-343
             Project: Singa
          Issue Type: New Feature
            Reporter: wangwei


Currently, all neural network operations including convolution, pooling, activation, etc are implemented in C++ Layers. It is not as flexible as providing them as Tensor functions, which can be combined to create new customized layers in Python.

This ticket is going to create neural net operations using C++, and replace the current layer implementation using the neural net operations.

For example, we create the conv2d and conv2d_backward operations to implement the Conv2D layer. conv2d and conv2d_backward can also be called in Python for autograd (SINGA-342) implementation.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)