Details
-
Task
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
None
Description
Since all our bias are column vector (which is consistent with Keras/Caffe), whereas bias of fully connected layer is a row-vector. Similarly, the weight that is passed to caffe is transpose of weights passed to SystemML (since both store in row-major NCHW format).
Making the dimensions consistent will simplify loading of Caffe/Keras models.