QASP: Query-Adaptive Robust Vector Search Policy
QASP aims to hit per-query recall targets with less wasted vector-search work.
The paper says fixed search settings hide large query-by-query recall swings behind average scores. QASP predicts a full recall progression curve before search, then derives the search depth needed for a requested recall target. The authors say it generalizes across targets, index settings, and datasets, with a reactive adjustment step that needs no extra inference. In experiments, they report 99% recall with 80% less data access and lower recall variance.
ArXiv · AI/CL/LG's note
The paper says fixed search settings hide large query-by-query recall swings behind average scores. QASP predicts a full recall progression curve before search, then derives the search depth needed for a requested recall target. The authors say it generalizes across targets, index settings, and datasets, with a reactive adjustment step that needs no extra inference. In experiments, they report 99% recall with 80% less data access and lower recall variance.
ArXiv · AI/CL/LG's note
score 4