Learning Poisson Binomial Distributions

TL;DR

This paper introduces efficient algorithms for learning Poisson Binomial Distributions, achieving near-optimal sample complexity and running time, and advances understanding of this fundamental distribution class.

Contribution

It provides the first highly efficient algorithms for learning PBDs with near-optimal sample complexity and running time, settling the complexity of this basic problem.

Findings

An algorithm learns PBDs with (rac{1}{\u03b5^3}) samples, independent of n.

A proper learning algorithm uses (rac{1}{^2}) samples with near-optimal complexity.

The algorithms are nearly optimal in sample complexity, matching lower bounds.

Abstract

We consider a basic problem in unsupervised learning: learning an unknown \emph{Poisson Binomial Distribution}. A Poisson Binomial Distribution (PBD) over $\{0,1,\dots,n\}$ is the distribution of a sum of $n$ independent Bernoulli random variables which may have arbitrary, potentially non-equal, expectations. These distributions were first studied by S. Poisson in 1837 \cite{Poisson:37} and are a natural $n$-parameter generalization of the familiar Binomial Distribution. Surprisingly, prior to our work this basic learning problem was poorly understood, and known results for it were far from optimal. We essentially settle the complexity of the learning problem for this basic class of distributions. As our first main result we give a highly efficient algorithm which learns to $\eps$-accuracy (with respect to the total variation distance) using $\tilde{O}(1/\eps^3)$ samples \emph{independent of $n$}. The running time of the algorithm is \emph{quasilinear} in the size of its input data, i.e., $\tilde{O}(\log(n)/\eps^3)$ bit-operations. (Observe that each draw from the distribution is a $\log(n)$-bit string.) Our second main result is a {\em proper} learning algorithm that learns to $\eps$-accuracy using $\tilde{O}(1/\eps^2)$ samples, and runs in time $(1/\eps)^{\poly (\log (1/\eps))} \cdot \log n$. This is nearly optimal, since any algorithm {for this problem} must use $\Omega(1/\eps^2)$ samples. We also give positive and negative results for some extensions of this learning problem to weighted sums of independent Bernoulli random variables.