MUVERA: Multi-Vector Retrieval via Fixed Dimensional Encodings

TL;DR

MUVERA introduces a novel method to convert multi-vector retrieval into single-vector retrieval using fixed dimensional encodings, enabling efficient and accurate information retrieval with theoretical guarantees and improved performance.

Contribution

The paper presents MUVERA, a new retrieval algorithm that approximates multi-vector similarity with single-vector encodings, reducing computational complexity and enabling use of standard MIPS solvers.

Findings

Achieves same recall as state-of-the-art heuristics with fewer candidates.

Retrieves 2-5 times fewer candidates for the same recall.

Improves end-to-end recall by 10% with 90% lower latency.

Abstract

Neural embedding models have become a fundamental component of modern information retrieval (IR) pipelines. These models produce a single embedding $x \in \mathbb{R}^d$ per data-point, allowing for fast retrieval via highly optimized maximum inner product search (MIPS) algorithms. Recently, beginning with the landmark ColBERT paper, multi-vector models, which produce a set of embedding per data point, have achieved markedly superior performance for IR tasks. Unfortunately, using these models for IR is computationally expensive due to the increased complexity of multi-vector retrieval and scoring. In this paper, we introduce MUVERA (MUlti-VEctor Retrieval Algorithm), a retrieval mechanism which reduces multi-vector similarity search to single-vector similarity search. This enables the usage of off-the-shelf MIPS solvers for multi-vector retrieval. MUVERA asymmetrically generates Fixed Dimensional Encodings (FDEs) of queries and documents, which are vectors whose inner product approximates multi-vector similarity. We prove that FDEs give high-quality $\epsilon$-approximations, thus providing the first single-vector proxy for multi-vector similarity with theoretical guarantees. Empirically, we find that FDEs achieve the same recall as prior state-of-the-art heuristics while retrieving 2-5$\times$ fewer candidates. Compared to prior state of the art implementations, MUVERA achieves consistently good end-to-end recall and latency across a diverse set of the BEIR retrieval datasets, achieving an average of 10$\%$ improved recall with $90\%$ lower latency.