Inferensys

Glossary

Vector OLE (VOLE)

A cryptographic primitive that allows two parties to generate a long vector of correlated oblivious linear evaluations, serving as a fast, low-communication foundation for recent high-performance PSI protocols.
Engineer reviewing vector database search results on laptop, embeddings visualization on screen, home office coding session.
CRYPTOGRAPHIC PRIMITIVE

What is Vector OLE (VOLE)?

Vector Oblivious Linear Evaluation (VOLE) is a cryptographic protocol enabling two parties to efficiently generate a long vector of correlated oblivious linear evaluations, serving as a fast, low-communication foundation for modern private set intersection (PSI) and secure computation.

Vector OLE (VOLE) is a two-party protocol where a sender holds a scalar x and a receiver holds a scalar y, and the parties securely compute a long vector of additive shares such that the receiver obtains (a_i, b_i) and the sender obtains (c_i, d_i) satisfying b_i = a_i * x + y + d_i. This generates a massive batch of correlated oblivious linear evaluations using only fast, symmetric-key operations after a brief base setup, dramatically reducing the amortized communication cost per evaluation compared to standard Oblivious Transfer (OT).

In high-performance Private Set Intersection (PSI) protocols like the KKRT and Ferret constructions, VOLE acts as the primary cryptographic engine for generating the correlated randomness needed to evaluate Oblivious Pseudorandom Functions (OPRFs). By replacing expensive public-key operations with VOLE-based OT extension, these protocols achieve near line-speed performance, making VOLE the critical primitive that enables practical, low-latency private contact discovery and secure data matching at scale.

Cryptographic Primitives

Key Features of VOLE

Vector Oblivious Linear Evaluation (VOLE) is a foundational cryptographic primitive that enables the rapid, low-communication generation of correlated randomness, serving as the performance engine behind modern, highly efficient Private Set Intersection (PSI) protocols.

01

The Core VOLE Correlation

VOLE is a two-party protocol where a sender inputs a scalar value x and a receiver inputs a scalar value y. The protocol outputs a random vector A and a scalar B to the sender, and a vector C to the receiver, such that the linear correlation C = A * x + B * y holds. Crucially, the sender learns nothing about y, and the receiver learns nothing about x or B, making it a fundamental building block for secure computation.

02

Random VOLE vs. Chosen-Input VOLE

VOLE can be instantiated in two primary modes:

  • Random VOLE: The inputs x and y are randomly generated by the protocol itself. This is the fastest form and is used to generate large batches of correlated randomness in a setup phase.
  • Chosen-Input VOLE: The parties provide their own private inputs x and y. This is the form directly consumed by applications like PSI, and it can be efficiently converted from random VOLE using a technique called oblivious transfer extension.
03

Foundation for High-Performance PSI

VOLE is the cryptographic engine that powers the world's fastest PSI protocols, such as those based on the KKRT and Ferret frameworks. By using VOLE to generate a massive number of oblivious pseudorandom function (OPRF) instances in bulk, these protocols can evaluate set membership with extremely low communication overhead. This replaces the slower, per-element public-key operations of older Diffie-Hellman-based PSI with fast, batched symmetric-key computations.

04

Efficient Instantiation via OT Extension

VOLE is not a standalone protocol but is itself constructed from more basic primitives. The most efficient constructions use Oblivious Transfer (OT) extension, specifically the IKNP protocol and its modern variants. A small number of base OTs are used as a seed to generate a virtually unlimited number of VOLE correlations using only fast symmetric-key operations, making the amortized cost per correlation negligible.

05

The Ferret Protocol and Quasi-Cyclic Codes

The Ferret OT protocol is a state-of-the-art method for generating VOLE correlations at blazing speed. It leverages quasi-cyclic low-density parity-check (LDPC) codes to achieve an extremely low communication overhead, approaching the theoretical minimum. This construction is a key reason why modern PSI protocols can process millions of items in seconds, even on consumer-grade hardware.

06

VOLE in the Random Oracle Model

The security of practical VOLE-based protocols is typically proven in the Random Oracle Model (ROM). Here, a hash function is treated as a truly random function. This allows for highly efficient constructions where the correlation-robustness of the hash function guarantees that the output vectors A and C are indistinguishable from random to the respective parties, preventing any leakage of the secret inputs x and y.

VECTOR OLE DEEP DIVE

Frequently Asked Questions

Explore the cryptographic primitive that powers the fastest modern Private Set Intersection protocols. These answers break down the mechanics, security, and performance of Vector Oblivious Linear Evaluation.

Vector Oblivious Linear Evaluation (VOLE) is a cryptographic protocol where a sender learns a random vector w and a scalar Δ, while a receiver learns a vector v and a scalar x, such that the correlation w = v + Δ · x holds over a finite field. Crucially, the sender remains oblivious to x, and the receiver remains oblivious to Δ.

The protocol works by generating a large batch of these correlated pairs efficiently. Modern constructions, like those based on the Learning Parity with Noise (LPN) assumption or using fast OT extension techniques, allow parties to generate millions of VOLE correlations with very low communication overhead. This makes VOLE a powerful, reusable resource for secure computation, acting as a fast, linear correlation factory.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.