# Sample Efficient Distribution Testing Using Centralized or Distributed Computation

**Date:**April 28, 2022

**Time:**4:00 pm

**Room:**DBH 4011

**Speaker:**Themis Gouleakis

**Abstract:**

In the first part of the talk, we study the problem of testing discrete distributions with a focus on the high probability regime. Specifically, given samples from one or more discrete distributions, a property $\mathcal{P}$, and parameters $0< \eps, \delta <1$, we want to distinguish with probability at least $1-\delta$ whether these distributions satisfy $\mathcal{P}$ or are $\eps$-far from $\mathcal{P}$ in total variation distance. Most prior work in distribution testing studied the constant confidence case (corresponding to $\delta = \Omega(1)$), and provided sample-optimal testers for a range of properties. While one can always boost the confidence probability of any such tester by black-box amplification, this generic boosting method typically leads to sub-optimal sample bounds.

Here we study the following broad question: For a given property $\mathcal{P}$, can we characterize the sample complexity of testing $\mathcal{P}$ as a function of all relevant problem parameters, including the error probability $\delta$? As our main results,

we provide the first algorithms for uniformity, closeness and independence testing that are sample-optimal, within constant factors, as a function of all relevant parameters. Our techniques naturally extend to give optimal testers for related problems.

Apart from the classical centralized computation model, we will also talk about distribution testing in a distributed and a streaming model with communication and memory constraints respectively. More specifically, we consider the following computational models: (1) The {\em one-pass streaming model} where the goal is to minimize the sample complexity of the protocol subject to a memory constraint, and (2) A {\em distributed model} where the data samples reside at multiple machines and the goal is to minimize the communication cost of the protocol. In these models, we will provide efficient algorithms for uniformity/identity testing (goodness of fit) and closeness testing (two sample testing).

Moreover, we show nearly-tight lower bounds on (1) the sample complexity

of any one-pass streaming tester for uniformity, subject to the memory constraint,

and (2) the communication cost of any uniformity testing protocol,

in a restricted “one-pass” model of communication.