Private and Online Learnability Are Equivalent

Noga Alon; Mark Bun; Roi Livni; Maryanthe Malliaris; Shay Moran

doi:10.1145/3526074

Private and Online Learnability Are Equivalent

Noga Alon, Mark Bun, Roi Livni, Maryanthe Malliaris, Shay Moran

Research output: Contribution to journal › Article › peer-review

Abstract

Let 'H be a binary-labeled concept class. We prove that 'H can be PAC learned by an (approximate) differentially private algorithm if and only if it has a finite Littlestone dimension. This implies a qualitative equivalence between online learnability and private PAC learnability.

Original language	English
Article number	28
Journal	Journal of the ACM
Volume	69
Issue number	4
DOIs	https://doi.org/10.1145/3526074
State	Published - 16 Aug 2022

Keywords

Differential privacy
Littlestone dimension
online learning
PAC learning

Access to Document

10.1145/3526074

Cite this

@article{725654511b3a4676aa6f7d6cd90a6946,

title = "Private and Online Learnability Are Equivalent",

abstract = "Let 'H be a binary-labeled concept class. We prove that 'H can be PAC learned by an (approximate) differentially private algorithm if and only if it has a finite Littlestone dimension. This implies a qualitative equivalence between online learnability and private PAC learnability.",

keywords = "Differential privacy, Littlestone dimension, online learning, PAC learning",

author = "Noga Alon and Mark Bun and Roi Livni and Maryanthe Malliaris and Shay Moran",

note = "Publisher Copyright: {\textcopyright} 2022 Association for Computing Machinery.",

year = "2022",

month = aug,

day = "16",

doi = "10.1145/3526074",

language = "אנגלית",

volume = "69",

journal = "Journal of the ACM",

issn = "0004-5411",

publisher = "Association for Computing Machinery (ACM)",

number = "4",

}

TY - JOUR

T1 - Private and Online Learnability Are Equivalent

AU - Alon, Noga

AU - Bun, Mark

AU - Livni, Roi

AU - Malliaris, Maryanthe

AU - Moran, Shay

PY - 2022/8/16

Y1 - 2022/8/16

N2 - Let 'H be a binary-labeled concept class. We prove that 'H can be PAC learned by an (approximate) differentially private algorithm if and only if it has a finite Littlestone dimension. This implies a qualitative equivalence between online learnability and private PAC learnability.

AB - Let 'H be a binary-labeled concept class. We prove that 'H can be PAC learned by an (approximate) differentially private algorithm if and only if it has a finite Littlestone dimension. This implies a qualitative equivalence between online learnability and private PAC learnability.

KW - Differential privacy

KW - Littlestone dimension

KW - online learning

KW - PAC learning

UR - http://www.scopus.com/inward/record.url?scp=85139218816&partnerID=8YFLogxK

U2 - 10.1145/3526074

DO - 10.1145/3526074

M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???

AN - SCOPUS:85139218816

SN - 0004-5411

VL - 69

JO - Journal of the ACM

JF - Journal of the ACM

IS - 4

M1 - 28

ER -

Private and Online Learnability Are Equivalent

Abstract

Keywords

Access to Document

Other files and links

Cite this