CConcentric Vector Chord

System design

Architecture overview

Conceptual pipeline for decentralized approximate nearest-neighbor search. Detailed algorithms and evaluation appear in the technical paper.

Query & placement flow

High-level pipeline from embedding insertion to approximate nearest-neighbor retrieval. This overview reflects the published algorithm without exposing implementation details.

Step 1

Embedding

Vector v in R^d

Step 2

Concentric key

Shell + angle to ring

Step 3

LSH / KLSH

Similarity bucket bits

Step 4

Chord owner

Successor + overflow

Step 5

k-NN query

Probe, score, top-k

What makes it different

  • Order-preserving concentric keys instead of uniform hashing
  • Optional kernelized LSH for implicit feature spaces
  • Hybrid search: LSH buckets + centroid filtering + rerank

Visualization legend

  • Blue ring — Chord identifier ring
  • Gray shells — embedding shells
  • Gold ticks — DHT nodes; orange path — query route