Sequential Mode Estimation with Oracle Queries

TL;DR

This paper develops adaptive algorithms for identifying the mode of a distribution using oracle queries, analyzing their efficiency and establishing lower bounds for different query models.

Contribution

It introduces sequential mode-estimation algorithms with proven query complexity bounds for two distinct oracle query models.

Findings

Algorithms achieve near-optimal query complexity.

Lower bounds match the algorithms' performance.

Effective strategies depend on the query model used.

Abstract

We consider the problem of adaptively PAC-learning a probability distribution $\mathcal{P}$'s mode by querying an oracle for information about a sequence of i.i.d. samples $X_1, X_2, \ldots$ generated from $\mathcal{P}$. We consider two different query models: (a) each query is an index $i$ for which the oracle reveals the value of the sample $X_i$, (b) each query is comprised of two indices $i$ and $j$ for which the oracle reveals if the samples $X_i$ and $X_j$ are the same or not. For these query models, we give sequential mode-estimation algorithms which, at each time $t$, either make a query to the corresponding oracle based on past observations, or decide to stop and output an estimate for the distribution's mode, required to be correct with a specified confidence. We analyze the query complexity of these algorithms for any underlying distribution $\mathcal{P}$, and derive corresponding lower bounds on the optimal query complexity under the two querying models.