Private Linear Computation: Algorithms, Performance Limits, and Applications to Machine Learning

This article surveys Private Computation, a foundational framework for achieving privacy across a broad range of applications. Private Computation techniques enable clients to perform computations on data stored at remote servers without revealing sensitive information about the computation, such as...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE BITS the information theory magazine s. 1 - 12
Hlavní autoři: Heidarzadeh, Anoosheh, Sprintson, Alex
Médium: Journal Article
Jazyk:angličtina
Vydáno: IEEE 14.11.2025
Témata:
Data privacy
Indexes
Information theory
Machine learning
Prediction algorithms
Privacy
Random variables
Servers
Upper bound
Vectors
ISSN:2692-4080, 2692-4110
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:This article surveys Private Computation, a foundational framework for achieving privacy across a broad range of applications. Private Computation techniques enable clients to perform computations on data stored at remote servers without revealing sensitive information about the computation, such as the specific function being computed or the identities of the data items involved in the computations. We focus on the Private Linear Computation (PLC) problem, where the goal is to privately retrieve a single linear combination of a subset of data items stored across remote servers. We begin by reviewing key information-theoretic privacy notions: functional privacy , which protects the coefficients of the desired linear combination; joint support privacy , which conceals the entire set of identities of the data items involved in the combination; and individual support privacy , which hides the identity of each individual data item involved. Next, we present PLC schemes for both single-server and multi-server scenarios, under these privacy notions. We also discuss the optimality of these schemes in terms of the total amount of information the client must download from the servers. Finally, we discuss the potential role of PLC schemes in enabling privacy-preserving statistical inference and machine learning on remote data.
ISSN:2692-4080
2692-4110
DOI:10.1109/MBITS.2025.3633196