Abstract
Random projection is often used to project higher-dimensional vectors onto alower-dimensional space, while approximately preserving their pairwisedistances. It has emerged as a powerful tool in various data processing tasksand has attracted considerable research interest. Partly motivated by therecent discoveries in neuroscience, in this paper we study the problem ofrandom projection using binary matrices with controllable sparsity patterns.Specifically, we proposed two sparse binary projection models that work ongeneral data vectors. Compared with the conventional random projection modelswith dense projection matrices, our proposed models enjoy significantcomputational advantages due to their sparsity structure, as well as improvedaccuracies in empirical evaluations.