Using Restricted Boltzmann Machines for Missing Data Imputation

Suman Ravuri (with Mike Seltzer of Microsoft Research)

ICSI

Tuesday, November 1, 2011
12:30 - 2:00pm

Restricted Boltzmann Machines (RBMs) have been used as a building block for Deep Belief Networks, but the characteristics of RBMs themselves have not been investigated. In this work, we investigate the performance of RBMs for missing data imputation of speech data. We will show how RBMs can be used for denoising speech signals using a novel update algorithm.