English

Learning $k$-Modal Distributions via Testing

Data Structures and Algorithms 2014-09-16 v3 Machine Learning Statistics Theory Statistics Theory

Abstract

A kk-modal probability distribution over the discrete domain {1,...,n}\{1,...,n\} is one whose histogram has at most kk "peaks" and "valleys." Such distributions are natural generalizations of monotone (k=0k=0) and unimodal (k=1k=1) probability distributions, which have been intensively studied in probability theory and statistics. In this paper we consider the problem of \emph{learning} (i.e., performing density estimation of) an unknown kk-modal distribution with respect to the L1L_1 distance. The learning algorithm is given access to independent samples drawn from an unknown kk-modal distribution pp, and it must output a hypothesis distribution p^\widehat{p} such that with high probability the total variation distance between pp and p^\widehat{p} is at most ϵ.\epsilon. Our main goal is to obtain \emph{computationally efficient} algorithms for this problem that use (close to) an information-theoretically optimal number of samples. We give an efficient algorithm for this problem that runs in time poly(k,log(n),1/ϵ)\mathrm{poly}(k,\log(n),1/\epsilon). For kO~(logn)k \leq \tilde{O}(\log n), the number of samples used by our algorithm is very close (within an O~(log(1/ϵ))\tilde{O}(\log(1/\epsilon)) factor) to being information-theoretically optimal. Prior to this work computationally efficient algorithms were known only for the cases k=0,1k=0,1 \cite{Birge:87b,Birge:97}. A novel feature of our approach is that our learning algorithm crucially uses a new algorithm for \emph{property testing of probability distributions} as a key subroutine. The learning algorithm uses the property tester to efficiently decompose the kk-modal distribution into kk (near-)monotone distributions, which are easier to learn.

Keywords

Cite

@article{arxiv.1107.2700,
  title  = {Learning $k$-Modal Distributions via Testing},
  author = {Constantinos Daskalakis and Ilias Diakonikolas and Rocco A. Servedio},
  journal= {arXiv preprint arXiv:1107.2700},
  year   = {2014}
}

Comments

28 pages, full version of SODA'12 paper, to appear in Theory of Computing

R2 v1 2026-06-21T18:36:28.496Z