Hierarchical Navigable Small-World graphs
PulseAugur coverage of Hierarchical Navigable Small-World graphs — every cluster mentioning Hierarchical Navigable Small-World graphs across labs, papers, and developer communities, ranked by signal.
14 day(s) with sentiment data
-
pg_pinecone brings vector search directly into PostgreSQL
Netics has released pg_pinecone, an open-source PostgreSQL extension that integrates vector search capabilities directly into the database. This extension offers three engines: a native engine that stores compressed vec…
-
New beam search variants enhance DiskANN for faster nearest-neighbor search
This paper introduces variants of beam search for DiskANN, a state-of-the-art method for large-scale approximate nearest-neighbor search. The research unifies and generalizes existing theoretical results on graph-based …
-
Developer builds local AI indexer to solve Claude's context memory issues
A developer has created Cortex, a local indexing tool for AI models like Claude, to address issues with context memory and data privacy. Cortex is a Go binary that indexes local files and integrates with AI coding assis…
-
Milvus vs. Qdrant: Production Architectures and Performance Deep Dive
This article provides a detailed comparison of Milvus and Qdrant, two prominent vector databases, focusing on their production architectures and performance. It highlights Milvus's decoupled storage and compute model, s…
-
REWEAVE method enhances vector database search with adaptive HNSW indexes
Researchers have developed REWEAVE, a novel online learning method for maintaining Hierarchical Navigable Small-World (HNSW) indexes used in vector databases. This technique adapts HNSW indexes to changing query workloa…
-
Handbook tackles RAG production pitfalls beyond tutorials
A new handbook addresses the gap between tutorial RAG systems and production-ready applications, highlighting common pitfalls in retrieval metrics, chunking strategies, and vector database costs. It emphasizes the impor…
-
New PiPNN methods drastically speed up ANNS index and k-NN graph construction
Researchers have developed PiPNN II and GPiPNN, significantly accelerating the construction of Approximate Nearest Neighbor Search (ANNS) indexes. PiPNN II achieves up to an 8x speedup on CPUs, while the GPU-implemented…
-
New Rust libraries boost ANN search for retrieval-augmented generation
Researchers have developed two open-source Rust libraries, Vectorium and kANNolo, designed to improve Approximate Nearest Neighbors (ANN) search for neural embeddings. These libraries offer a modular approach, separatin…
-
Qdrant 1.20 integrates BM25 keyword ranking into vector database index
Qdrant has released version 1.20, introducing significant enhancements to its open-source vector database. This update integrates BM25 keyword ranking directly into the full-text index, eliminating the need for separate…
-
GraphRAG enhances enterprise document retrieval over vector search
A new research paper proposes GraphRAG, a novel approach to retrieval-augmented generation (RAG) that utilizes knowledge graphs to enrich enterprise document retrieval. In experiments using a snapshot of the SEC EDGAR d…
-
AI API reliability and vector database optimization discussed
A technical discussion explores optimizing vector databases for AI workloads, specifically how Schemity's pgvector implementation handles title edits and Hierarchical Navigable Small World (HNSW) index growth. The conve…
-
pgvector HNSW filtering bug causes RAG retrieval failures
A user encountered significant issues with pgvector's HNSW index when implementing a retrieval-augmented generation (RAG) system for a multi-tenant documentation assistant. The core problem was that PostgreSQL's filteri…
-
New HNSW compression technique cuts graph size by 50% with no latency penalty
Researchers have developed a novel lossless compression technique for Hierarchical Navigable Small World (HNSW) graphs, a crucial component in modern vector search systems. This method, which involves offline graph perm…
-
GraphRAG enhances enterprise document retrieval over vector RAG
Researchers have developed a new approach called GraphRAG to enhance retrieval-augmented generation (RAG) systems for enterprise document retrieval. By integrating a knowledge graph with metadata and frequently changing…
-
Nexariza AI unveils hybrid RAG for zero-hallucination enterprise knowledge engines
Nexariza AI is developing a new approach to Retrieval-Augmented Generation (RAG) for enterprise clients, aiming to eliminate hallucinations and ensure data sovereignty. Their system combines dense vector search with spa…
-
Holo-NN: New algebraic framework for private nearest neighbor search
Researchers have introduced Holo-NN, a novel algebraic framework for privacy-preserving approximate nearest neighbor (ANN) search in high-dimensional embeddings. This method replaces traditional branching operations, wh…
-
Deep Fusion Memory boosts LLM accuracy and speed with integrated knowledge retrieval
A new architecture called Deep Fusion Memory (DFM) embeds chunk-level parametric memory directly into a language model's weights, offering a hybrid approach to knowledge retrieval. This method aims to improve LLM accura…
-
RAG chatbot errors traced to retrieval, not LLM; fixes detailed
Retrieval-augmented generation (RAG) chatbots can provide incorrect or fabricated answers due to issues in the retrieval or generation phases, rather than the language model itself. Common problems include failing to fi…
-
Lychee research paper introduces faster, high-recall vector search
A new research paper introduces Lychee, a GPU-accelerated inverted-file search system designed for high-recall vector search. Lychee achieves significant speedups over existing methods like IVF-Flat, CAGRA, and HNSW by …
-
pgvector HNSW index struggles with concurrent writes and vector normalization
A developer encountered significant performance issues with pgvector's Hierarchical Navigable Small World (HNSW) index when handling concurrent writes, specifically during the integration of an AI job search pipeline. T…