AI Paper Atlas

Research map / Graph learning / Graph algorithms and clustering

DiskANN: Fast Accurate Billion-point Nearest Neighbor Search on a Single Node

Suhas Jayaram Subramanya, Fnu Devvrit, Harsha Vardhan Simhadri, Ravishankar Krishnawamy, Rohan Kadekodi

NeurIPS 2019 · Graph learning › Graph algorithms and clustering · approach: Graph distance and retrieval · 46 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
single one-shot Vamana index replaced by merged index of overlapping shards (ablation)DiskANN on SIFT1B · SIFT1B bigannsearch latency at target recall: up to 20% extra latencyHurt
merged index enables building under 64GB RAM instead of ~1100GB (ablation)Vamana on SIFT1B · SIFT1B bigannpeak memory during build: ~1100GB -> under 64GBImproved
Vamana with alpha > 1 versus HNSW and NSG (alpha = 1)graph index · SIFT1M-scale, 98% 5-recall@5hops to converge: 2-3x fewer hopsImproved
Vamana indexing versus HNSW and NSG for in-memory searchgraph index · SIFT1M, GIST1M, DEEP1Mrecall vs latencyImproved
Vamana indexing versus HNSW and NSGDEEP1M indexing · DEEP1Mindex build time: 149s vs 219s (HNSW) vs 480s (NSG)Improved
add a second indexing pass with alpha >= 1 over the dataset (ablation)Vamana · graph constructiongraph qualityImproved
initialize from random graph instead of empty graph (ablation)Vamana · graph constructiongraph qualityImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy