Statistical properties of sketching algorithms

TL;DR

This paper analyzes the statistical properties of sketching algorithms used for data compression in large datasets, deriving new theoretical results and demonstrating their practical implications for linear regression.

Contribution

It introduces a statistical framework for understanding sketching algorithms, deriving distributional results and a central limit theorem for sketched estimators.

Findings

Conditional CLT for data-oblivious sketches

Optimal sketching algorithm depends on signal-to-noise ratio

Theoretical insights validated on real datasets

Abstract

Sketching is a probabilistic data compression technique that has been largely developed in the computer science community. Numerical operations on big datasets can be intolerably slow; sketching algorithms address this issue by generating a smaller surrogate dataset. Typically, inference proceeds on the compressed dataset. Sketching algorithms generally use random projections to compress the original dataset and this stochastic generation process makes them amenable to statistical analysis. We argue that the sketched data can be modelled as a random sample, thus placing this family of data compression methods firmly within an inferential framework. In particular, we focus on the Gaussian, Hadamard and Clarkson-Woodruff sketches, and their use in single pass sketching algorithms for linear regression with huge $n$. We explore the statistical properties of sketched regression algorithms and derive new distributional results for a large class of sketched estimators. A key result is a conditional central limit theorem for data oblivious sketches. An important finding is that the best choice of sketching algorithm in terms of mean square error is related to the signal to noise ratio in the source dataset. Finally, we demonstrate the theory and the limits of its applicability on two real datasets.