Caffe uses force_backward=True to allow one to compute gradients for data layers to do things such as DeepDream. In Mocha, you have empty implementations of backwards for layers such as data and crop. With something like the _forcebackward option, one could optionally compute diffs.
Caffe uses
force_backward=Trueto allow one to compute gradients for data layers to do things such as DeepDream. In Mocha, you have empty implementations of backwards for layers such as data and crop. With something like the _forcebackward option, one could optionally compute diffs.