tranpose parameter added to IP layer to support tied weights in an autoencoder. Argum...
authorYoussef Kashef <youssef.kashef@gmail.com>
Fri, 29 Jan 2016 18:21:48 +0000 (19:21 +0100)
committerYoussef Kashef <youssef.kashef@gmail.com>
Sat, 20 Feb 2016 14:40:05 +0000 (15:40 +0100)
commit8f847fa8fae0460c6bf8e8d7a9bcf96a44305033
tree7d341989ab034a1846476a6b40be114db608ea5a
parenteac9dd8145cb4b2a831046e1ddc024f8b865a2f9
tranpose parameter added to IP layer to support tied weights in an autoencoder. Arguments to matrix multiplication function are conditioned on this parameter, no actual transposing takes place.

test ip gradient computation with transpose on
include/caffe/layers/inner_product_layer.hpp
src/caffe/layers/inner_product_layer.cpp
src/caffe/layers/inner_product_layer.cu
src/caffe/proto/caffe.proto
src/caffe/test/test_inner_product_layer.cpp