Similar Vector Querying Using Quantized Central and Residual Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing devices lack specificity and optimization for efficient similarity querying of vectors, leading to low processing efficiency.
Innovation Solution
A method and apparatus that quantify candidate vectors into central and residual vector portions, allowing for pre-calculation and storage of common and pre-calculation results, which are then used to determine similar candidate vectors without repeated calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If general-purpose computing devices (CPU/GPU) are used for vector similarity querying, then the system can handle general computing requirements, but processing efficiency is low due to lack of hardware optimization for specific similarity querying operations
Solution Approach 1:
The patent segments the vector similarity computation into distinct operational phases: coarse filtering using quantized vector portions, followed by precise similarity calculation using residual vector portions. This segmentation allows general-purpose devices to handle different computation stages efficiently, improving overall processing throughput without requiring specialized hardware for each operation type.
Solution Approach 2:
The patent performs preliminary quantization of candidate vectors into central and residual portions before the actual similarity search. This pre-processing step enables faster coarse filtering in the first stage, reducing the number of vectors that require computationally intensive precise calculation, thereby improving processing efficiency on general-purpose devices.
2Measurement precision
If precise similarity calculation is performed for all candidate vectors, then high query accuracy is achieved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent applies partial action by performing precise similarity calculation only on a subset of candidate vectors that pass the coarse filtering stage. Instead of calculating exact similarity for all vectors in the database, the system first filters using quantized portions and then applies precise calculation only to promising candidates, significantly reducing computational complexity while maintaining high query accuracy.
Solution Approach 2:
The patent substitutes the mechanical computation of full vector similarity with a two-stage computational approach: first using simplified distance calculation on quantized portions, then using precise calculation only when necessary. This substitution reduces the overall computational burden while preserving measurement precision for the final results.
Data Source
AI summary
A method for querying in a candidate vector set candidate vectors similar to object vectors is disclosed, wherein the candidate vector set comprises a plurality of candidate vectors each being quantized as having a central vector portion and a residual vector portion, the method comprising: querying, a first number of candidate vector subsets that are closest to the object vector; generating and storing a plurality of common calculation results based on a set of central vector portions and a set of residual vector portions of candidate vectors of the first number of candidate vector subsets; generating and storing pre-calculation results based on the set of object vectors and the set of residual vector portion; and determining, a second number of candidate vectors that are similar to each object vector in the corresponding first number of candidate vector subsets based on the stored pre-calculation results and common calculation results.