Vector Query Accuracy Calibration for Search Latency Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vector indexes do not allow users to specify a target accuracy in their respective queries, and existing vector index technologies fail to provide users with the desired level of accuracy and response time, leading to inefficient and inaccurate search results.
Innovation Solution
A system and method for processing vector queries that allow users to specify a target accuracy value, which adjusts vector search parameters such as the number of probes in IVF or the size of the Top-K heap in HNSW, optimizing search results based on user-defined accuracy requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If vector index search uses approximate indexing techniques to improve search speed, then response time is reduced, but search accuracy deteriorates
Solution Approach 1:
The patent applies dynamics by making search parameters adjustable and adaptive. The system allows users to dynamically specify target accuracy levels and automatically adjusts search parameters (such as number of probes in IVF or efSearch in HNSW) based on the requested accuracy, enabling the search system to adapt between speed and accuracy requirements in real-time
Solution Approach 2:
The patent changes physical or computational parameters of the vector index search process. By mapping target accuracy values to specific search parameter values (such as changing the number of centroid partitions to scan or the size of the heap to maintain), the system achieves different accuracy levels while maintaining acceptable response times
2Measurement precision
If vector index search increases the number of probes or search depth to improve accuracy, then search accuracy is improved, but response time increases
Solution Approach 1:
The patent changes search parameters such as the number of probes in IVF or the efSearch parameter in HNSW based on the target accuracy requirement. By establishing mappings between accuracy levels and parameter values, the system optimizes the balance between accuracy and time without requiring full exhaustive searches
Solution Approach 2:
The patent applies partial action by performing only the necessary amount of search work required to achieve the target accuracy. Instead of always performing complete exhaustive searches, the system performs partial searches (fewer probes) when lower accuracy is acceptable, reducing time while maintaining sufficient accuracy for the application
3Ease of operation
If vector index uses fixed search parameters to simplify system operation, then ease of operation is improved, but adaptability to different accuracy requirements deteriorates
Solution Approach 1:
The patent applies self-service by having the system automatically adjust search parameters based on user-specified accuracy requirements. Instead of requiring manual configuration of complex search parameters, users simply specify their accuracy needs and the system automatically selects appropriate parameters through pre-established mappings, maintaining simplicity while enabling adaptability
Solution Approach 2:
The patent makes the search system universal by enabling it to handle different accuracy requirements through a single interface. The same search system can serve applications with different accuracy needs by automatically adapting parameters based on the requested accuracy level, eliminating the need for multiple specialized search systems
Data Source
AI summary
Techniques for automatically calibration the accuracy of vector queries is provided. In one technique, a vector query that includes a query vector and that is associated with an accuracy value is received. The accuracy value may be a percentage value. In response to receiving the vector query, a value for a vector search parameter is determined based on the accuracy value and a plurality of past accuracy scores. For IVF vector indexes, the vector search parameter may be a number of centroid partitions to scan during the search. For HNSW vector indexes, the vector search parameter value may be a size of a results heap. A search of a vector index is performed based on the query vector and the value for the vector search parameter. A set of results is generated based on the search of the vector index.