Layer-wise learning of deep generative models

Ludovic Arnold, Yann Ollivier

When using deep, multi-layered architectures to build generative models of data, it is difficult to train all layers at once. We propose a layer-wise training procedure admitting a…