General and Practical Tuning Method for Off-the-Shelf Graph-Based Index: SISAP Indexing Challenge Report by Team UTokyo
This provides a practical tuning method for users of graph-based ANN indexes, though it is incremental as it optimizes existing systems rather than creating new ones.
The paper tackles the problem of optimal tuning for off-the-shelf graph-based approximate nearest neighbor indexes by introducing a black-box optimization method that adjusts parameters like dimension, database size, and entry points to meet recall and QPS targets. It achieved second place in the SISAP 2023 Indexing Challenge's 10M and 30M tracks, showing substantial performance improvements over brute-force methods.
Despite the efficacy of graph-based algorithms for Approximate Nearest Neighbor (ANN) searches, the optimal tuning of such systems remains unclear. This study introduces a method to tune the performance of off-the-shelf graph-based indexes, focusing on the dimension of vectors, database size, and entry points of graph traversal. We utilize a black-box optimization algorithm to perform integrated tuning to meet the required levels of recall and Queries Per Second (QPS). We applied our approach to Task A of the SISAP 2023 Indexing Challenge and got second place in the 10M and 30M tracks. It improves performance substantially compared to brute force methods. This research offers a universally applicable tuning method for graph-based indexes, extending beyond the specific conditions of the competition to broader uses.