YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Barycentric Simplicial Hashing (BSH) πŸ”Ί

A Four-State Topological Hash Competitive with Industry-Standard Product Quantization at Half the Memory.

Author: Andres Sebastian Pirolo
ORCID: 0009-0004-3899-1222
Email: lctrnc1@gmail.com
DOI: [insert after Zenodo upload]


πŸš€ The Breakthrough

Barycentric Simplicial Hashing (BSH) is a hardware and software-agnostic mathematical framework that uses the local triangulation of a vector space as a discrete coordinate system for Approximate Nearest Neighbor (ANN) search.

By extending the natural three-vertex zone assignment with a fourth state (the triangle barycentre), BSH creates a topological buffer zone. This turns the curse of dimensionality into a feature, assigning >51% of vectors to the barycentre region in 24D spaces.

πŸ“Š Core Benchmark (Out-of-Sample)

Method Recall@1 (OOS) Memory Footprint Lift over Random
3-state Hash 64.5% 34 bytes 14.3x
BSH 4-state (This Work) 84–90% 34–57 bytes 18.7x
FAISS IVF-PQ 85–90% 64 bytes β€”

Tested on completely independent distributions (never seen during index construction) using ARM Cortex-X3.


πŸ”’ Intellectual Property & Commercial Licensing

This repository contains the official research paper outlining the mathematical algorithms and pseudocode for BSH.

The theoretical framework is published as Prior Art to prevent third-party patenting of simplicial complex indexing, four-state barycentric quantization, and related RAG/LLM applications.

Proprietary Implementation Available

The highly optimized, production-ready C++ / ARM NEON SIMD implementation is retained as a Trade Secret. It is exceptionally fast, avoiding full matrix instantiations via implicit L1 Laplacian power iterations and barycentric popcount hashing.

Strategic Commercial Licensing & First-Mover Advantage: Entities seeking to integrate BSH into production LLM inference, vector databases, or on-device AI accelerators should secure licensing rights.

For commercial licensing of the intellectual property or the proprietary C++ engine, contact: lctrnc1@gmail.com


πŸ“„ License

  • Paper: CC BY-NC-ND 4.0
  • Commercial Application of the Math/Algorithms: Requires explicit written authorization. See LICENSE.md for full Layer-3 Commercial Notice.
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support