Advances in Training Restricted Boltzmann Machines

Recently deep learning has become a field of attention in machine learning (see [1, 2] for a review). Deep learning mainly focuses on applying a deep neural network to various machine learning applications where there are abundant amount of highdimensional training samples either labeled or unlabeled. Especially, it was shown that deep models are able to extract efficient, hierarchical features. Many recent advances in deep learning have shown performances that are far superior to conventional approaches in many areas including image and speech recognition as well as multi-modal learning.