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.
9 day(s) with sentiment data
-
AI models spark debate on software engineering's future, while benchmarks reveal performance metrics · 3 sources tracked
Recent discussions in the AI community are debating the impact of new models like Fable and GPT-Astra on software engineering roles. Simultaneously, technical benchmarks are emerging, such as a 3.5 ms NumPy search over …
-
Vector Databases: A Deep Dive into RAG Indexing, Hybrid Search, and Scaling
This article delves into the technical aspects of implementing retrieval-augmented generation (RAG) systems, focusing on the crucial role of vector databases. It explores various indexing techniques such as Hierarchical…
-
GraphRAG faces security scare as new vulnerabilities emerge
Recent developments in Retrieval-Augmented Generation (RAG) highlight both efficiency gains and emerging security concerns. Several arXiv papers explore cheaper alternatives to traditional GraphRAG, such as the Matryosh…
-
pgvector HNSW filtering bug causes RAG systems to return zero results
A developer encountered a critical issue with retrieval-augmented generation (RAG) systems using the pgvector extension in PostgreSQL. The problem stems from how pgvector's Hierarchical Navigable Small World (HNSW) inde…
-
ModelLakeFishing framework enables efficient retrieval from million-scale model lakes
Researchers have developed ModelLakeFishing, a novel framework designed to efficiently retrieve suitable models from vast collections, known as model lakes, which can contain millions of reusable models. This system con…
-
pgvector extension brings vector search to PostgreSQL, offering cost savings
The pgvector extension for PostgreSQL offers a way to store and query vector embeddings directly within an existing relational database. This approach leverages PostgreSQL's existing infrastructure for ACID compliance, …
-
SOLO index offers certified recall for similarity search with reduced memory
Researchers have introduced SOLO, a novel index for approximate nearest-neighbor search in metric spaces that offers certified recall without requiring heuristic ranking. This method computes recall directly from the in…
-
LLM-Guided Pruning Enhances Nearest Neighbor Search Indices
Researchers have developed a new framework called LLM-Guided Graph Pruning (LGP) to improve the performance of approximate nearest neighbor search (ANNS) indices. This method uses Large Language Models (LLMs) to refine …
-
RAG pipeline speeds up 3.2x by switching vector index to HNSW
A Mastodon user reported a significant performance improvement in their retrieval-augmented generation (RAG) pipeline. By switching from FAISS IVF-Flat to Hierarchical Navigable Small World graphs (HNSW), they achieved …
-
HNSW vector search can silently miss data; tune ef_search
Vector search systems using Hierarchical Navigable Small World (HNSW) graphs can silently miss relevant data due to their approximate nature. The `ef_search` parameter, which controls the size of the candidate list duri…
-
Quantum Granular-Ball Learning Enhances ML Efficiency and Robustness
Two new research papers introduce Quantum Granular-Ball Learning (QGB-W$k$NN) and Granular-Ball Quantum Clustering (GBQC) frameworks. These methods aim to improve the efficiency and robustness of machine learning tasks,…
-
Hybrid Search Explained: Combining BM25 and Vector Embeddings for AI
This article explains the concept of hybrid search, which combines traditional keyword-based search (like BM25) with modern vector search. Vector search uses embeddings to represent text as numerical vectors, allowing f…
-
New In-Browser SQL Database Zeta-Lite Enhances AI Agent Memory
Researchers have developed Zeta-Lite, a new in-browser SQL database engine designed for AI agents. This WebAssembly-based database compiles the Zeta engine into a compact artifact, offering capabilities previously unava…
-
New research reveals Sublinear Power Law in vector search scalability
A new paper published on arXiv introduces the "Sublinear Power Law" to describe the scalability of graph-based vector search. Researchers found that search cost grows as N^c (where c is less than 1) when dataset size (N…
-
New research predicts ANN search performance from embedding statistics
A new research paper introduces methods to predict the performance of approximate nearest neighbor (ANN) search indexes based on embedding statistics. The paper demonstrates that index behavior, such as recall rates, ca…
-
New disk-based vector database AsterVec uses less RAM for local AI
The developer of AsterVec has created a new disk-based vector database designed to operate within a configurable memory budget, addressing the RAM competition between local LLMs and embedded vector stores. Unlike tradit…
-
Custom HNSW implementation slower than brute-force search in benchmarks
A developer built and benchmarked a retrieval engine from scratch, comparing a custom Hierarchical Navigable Small World (HNSW) implementation against the Faiss library. Surprisingly, brute-force search methods outperfo…
-
New research reveals fundamental limits in RAG defenses against coordinated poisoning attacks
Researchers have demonstrated a fundamental limitation in current defenses against coordinated poisoning attacks on vector retrieval systems used in retrieval-augmented generation (RAG). These admission-time defenses, w…
-
Vector database admission control tackles retrieval hubs under workload drift
Researchers have developed a new admission control mechanism for vector databases to mitigate the impact of retrieval hubs, where a single document dominates search results. This system maintains a set of sentinel queri…
-
Cerebras Knowledge Base Evolves with MCP Server and Refined Retrieval
This series of posts details the development of a knowledge base system for Cerebras, focusing on its retrieval and agent capabilities. Initially, the system used a hybrid retrieval method with an LLM reranker, achievin…