Published on Wed Feb 25 2015

### On Convolutional Approximations to Linear Dimensionality Reduction Operators for Large Scale Data Processing

,

In this paper, we examine the problem of approximating a general linear dimensionality reduction (LDR) operator. We establish a fundamental result, that most large LDR matrices in fact cannot be well approximated by partial circulant matrices. We then propose a natural

0
0
0
###### Abstract

In this paper, we examine the problem of approximating a general linear dimensionality reduction (LDR) operator, represented as a matrix $A \in \mathbb{R}^{m \times n}$ with , by a partial circulant matrix with rows related by circular shifts. Partial circulant matrices admit fast implementations via Fourier transform methods and subsampling operations; our investigation here is motivated by a desire to leverage these potential computational improvements in large-scale data processing tasks. We establish a fundamental result, that most large LDR matrices (whose row spaces are uniformly distributed) in fact cannot be well approximated by partial circulant matrices. Then, we propose a natural generalization of the partial circulant approximation framework that entails approximating the range space of a given LDR operator over a restricted domain of inputs, using a matrix formed as a product of a partial circulant matrix having rows and a 'post processing' matrix. We introduce a novel algorithmic technique, based on sparse matrix factorization, for identifying the factors comprising such approximations, and provide preliminary evidence to demonstrate the potential of this approach.