mirror of
https://github.com/allaunthefox/Research-Stack.git
synced 2026-08-20 15:37:29 +00:00
7e3858d88d
Orthogonal tensor (hypercube) assumes independent axes. Shear into parallelotope (hyper-rhomboid) models entangled dimensions. The shear angle encodes correlation strength; the Gram matrix of the shear IS the compression dictionary. 6 stack mappings: - PIST n-D: Cartesian → Bundle → Radial = hypercube → rhomboid → collapsed - Topological state machine: transition = shear on state tensor - N-D Gene Hypothesis: gene = n-D rhomboid, 3D structure = projection shadow - FAMM: preshaped delay = sheared time-domain rhomboid - OAC: latent cavity in sheared rhomboid space - Waveprobe: curvature = local shear angle of coordinate basis 3 compression interpretations + information gravity metric tensor
16 lines
943 B
Text
16 lines
943 B
Text
created: 20260507000000000
|
|
modified: 20260507000000000
|
|
tags: ResearchStack Search ANN Vector Compression
|
|
title: HNSW Vector Search
|
|
type: text/vnd.tiddlywiki
|
|
|
|
! HNSW Vector Search
|
|
|
|
Planned HNSW (Hierarchical Navigable Small World) upgrade for the semantic search pipeline. Currently [[Swarm ENE Middleware]] uses O(N) brute-force cosine similarity over the `swarm_semantic_index` table. HNSW provides O(log N) approximate nearest neighbor search with M=16 max connections, ef_construction=200, using cosine distance. Will index all 14D concept vectors from cached queries. Cold-start fallback to brute force. Target: <1ms search on 10k+ vectors with >95% recall at k=10. Referenced in the [[ENe Cognitive Refactor Plan]], Phase 6. Also underpins the [[Semantic Graph Mining]] pipeline and [[Equation Forest Index]] retrieval.
|
|
|
|
!! Links
|
|
|
|
* [[Swarm ENE Middleware]]
|
|
* [[Concept Vector 14]]
|
|
* [[Semantic Graph Mining]]
|
|
* [[Equation Forest Index]]
|