EMA: Approximate Nearest Neighbor Search with General Attribute Filtering and Dynamic Updates

Mocheng Li, Baotong Lu, James Cheng, Chenhao Ma · 2026

Read EMA: Approximate Nearest Neighbor Search with General Attribute Filtering and Dynamic Updates by Mocheng Li, Baotong Lu, James Cheng, Chenhao Ma (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-00734.txt.

Description

Li and co-authors speed up search in vector databases in this 2026 paper. The task is filtered nearest neighbor search: find the closest vectors that also pass a set of rules. Such search backs shopping advice, chat tools that fetch facts, and agent memory. Most past methods handle only range or label rules, and they cost much time and memory to build.

Their method, EMA, takes multi-part rules over both number and category fields. It also handles live updates well. Small summaries called Markers sit on graph edges and guide the search. Markers blend rule checks with shape cues and never wrongly rule a match out. A joint search then walks the graph with a bounded fix-up step for lost edges. The authors report speedups from 1.68 to 12.25 times over top general methods across varied workloads.

Catalog record

CREATORMocheng Li, Baotong Lu, James Cheng, Chenhao Ma
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.00734

Browse the full library · About the Tower of Babel