Fixed-Dimensional Query Encoding for Faster Multi-Vector Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multi-vector retrieval systems are computationally expensive and lack efficient algorithms with provable approximation guarantees, limiting their application in information retrieval tasks.
Innovation Solution
A multi-vector retrieval system that transforms multi-vector representations into fixed dimensional encodings (FDEs), enabling single-vector similarity search with provable approximation guarantees, using a Multi-Vector Retrieval Algorithm (MUVERA) to reduce computational cost and improve efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-vector retrieval is performed using traditional methods, then retrieval accuracy is improved, but computational cost and latency increase significantly
Solution Approach 1:
The patent transforms the multi-vector retrieval problem by changing the parameter representation from multiple vectors to a single fixed-dimensional encoding vector. This parameter transformation allows the system to maintain retrieval accuracy while enabling the use of efficient single-vector search algorithms, thereby reducing computational cost and latency significantly.
Solution Approach 2:
The patent extracts the essential information from multiple embedding vectors and consolidates it into a single fixed-dimensional encoding vector. This extraction process removes the computational complexity of handling multiple vectors while preserving the critical similarity information needed for accurate retrieval.
2Measurement precision
If multi-vector representations are used for detailed information capture, then retrieval quality is improved, but system complexity increases
Solution Approach 1:
The patent merges multiple embedding vectors into a single fixed-dimensional encoding vector through a consolidation process. This merging operation simplifies the system architecture by reducing the number of vectors from multiple to one, while maintaining the ability to capture detailed information through the encoding transformation.
3Measurement precision
If exact k-nearest neighbors search is performed on multi-vectors, then retrieval precision is improved, but computational resources are exhausted
Solution Approach 1:
The patent changes the parameter dimension from multi-vector to single fixed-dimensional vector, which enables the use of efficient single-vector k-NN search algorithms. This parameter transformation maintains retrieval precision while dramatically reducing the computational resources required for the search operation.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for multi-vector retrieval via fixed dimensional encodings. In one aspect, a method includes: obtaining a set of embedding vectors of a query in an embedding vector space; obtaining an encoded dataset including, for each data item in a set of data items, a respective encoded vector of the data item in a target vector space; encoding the set of embedding vectors of the query in the embedding vector space into an encoded vector of the query in the target vector space; performing, with respect to the encoded vector of the query, a k-nearest neighbors search on the respective encoded vectors of the data items in the encoded dataset; and identifying, from the k-nearest neighbors search, a top-k subset of the set of data items.