Question about the convEncoder
wenhuchen opened this issue · comments
In the file lvsr/bricks/__init__.py, the apply function of class ConvEncoder looks a bit weird. It seems to add the kernel function and do a summation over a window to get the output, it's quite different from the normal spatialconvolution function. Do I understand correctly?
ConvEncoder was an attempt to reproduce the results by Ranzato et. al. It is described in subsection 6.2 of the manuscript. It is indeed a very restricted architecture.