In summer 2017, I was an intern at the Purdue University working under Prof Bruno Ribeiro on improving the training of Restricted Boltzmann Machines. We used the Las Vegas transformation of Markov Chain Monte Carlo method to obtain better samples to estimate the negative phase of the gradient. The model trained via this method achieved a significantly higher likelihood on the MNIST data as compared to the conventional model trained via Contrastive Divergence. This repository contains a brief report on my work.
In summer 2017, I was an intern at the Purdue University working under Prof Bruno Ribeiro on improving the training of Restricted Boltzmann Machines. We used the Las Vegas transformation of Markov Chain Monte Carlo method to obtain better samples to estimate the negative phase of the gradient. The model trained via this method achieved a significantly higher likelihood on the MNIST data as compared to the conventional model trained via Contrastive Divergence. This repository contains a brief report on my work.