Learning poisson binomial distributions
We consider a basic problem in unsupervised learning: learning an unknown \emph{Poisson Binomial Distribution}. A Poisson Binomial Distribution (PBD) over {0,1,…,n} is the distribution of a sum of n independent Bernoulli random variables which may have arbitrary, potentially non-equal, expectations....
| Main Authors: | , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Association for Computing Machinery (ACM)
2021
|
| Online Access: | https://hdl.handle.net/1721.1/137414 |
| _version_ | 1826199204787126272 |
|---|---|
| author | Daskalakis, Constantinos Diakonikolas, Ilias Servedio, Rocco A. |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Daskalakis, Constantinos Diakonikolas, Ilias Servedio, Rocco A. |
| author_sort | Daskalakis, Constantinos |
| collection | MIT |
| description | We consider a basic problem in unsupervised learning: learning an unknown \emph{Poisson Binomial Distribution}. A Poisson Binomial Distribution (PBD) over {0,1,…,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. |
| first_indexed | 2024-09-23T11:16:06Z |
| format | Article |
| id | mit-1721.1/137414 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T11:16:06Z |
| publishDate | 2021 |
| publisher | Association for Computing Machinery (ACM) |
| record_format | dspace |
| spelling | mit-1721.1/1374142022-09-27T18:19:14Z Learning poisson binomial distributions Daskalakis, Constantinos Diakonikolas, Ilias Servedio, Rocco A. Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory We consider a basic problem in unsupervised learning: learning an unknown \emph{Poisson Binomial Distribution}. A Poisson Binomial Distribution (PBD) over {0,1,…,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. 2021-11-05T11:05:38Z 2021-11-05T11:05:38Z 2012 2019-05-15T17:55:03Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/137414 Daskalakis, Constantinos, Diakonikolas, Ilias and Servedio, Rocco A. 2012. "Learning poisson binomial distributions." en 10.1145/2213977.2214042 Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf Association for Computing Machinery (ACM) arXiv |
| spellingShingle | Daskalakis, Constantinos Diakonikolas, Ilias Servedio, Rocco A. Learning poisson binomial distributions |
| title | Learning poisson binomial distributions |
| title_full | Learning poisson binomial distributions |
| title_fullStr | Learning poisson binomial distributions |
| title_full_unstemmed | Learning poisson binomial distributions |
| title_short | Learning poisson binomial distributions |
| title_sort | learning poisson binomial distributions |
| url | https://hdl.handle.net/1721.1/137414 |
| work_keys_str_mv | AT daskalakisconstantinos learningpoissonbinomialdistributions AT diakonikolasilias learningpoissonbinomialdistributions AT servedioroccoa learningpoissonbinomialdistributions |