hnsw

Palette Atlas 用 8×16 维 Sliced Wasserstein 嵌入加手写 HNSW，在 48,695 幅公有领域画作上实现按配色检索：L1 距离对精确 EMD 的 top-10 重合率达 78%，recall@10 在 ef=32 时达 1.000、延迟仅 0.25ms（暴力搜索需 14ms）。

This article presents vector-bench, a tool designed to benchmark vector indexes across different databases with consistent conditions to accurately measure approximate nearest neighbor search performance.

The author built a retrieval engine from scratch to benchmark HNSW against FAISS, finding that brute force search is faster for small document sets and that encoder latency dominates retrieval time, while RRF fusion of BM25 and dense retrieval improves quality significantly.

The article discusses ACORN's algorithm for fixing filtered vector search challenges and compares it with Qdrant's filterable HNSW method, including performance benchmarks on a million-point dataset.

HyphaeDB is a novel agent-native memory infrastructure that repurposes the HNSW graph topology as a gossip-based communication fabric for multi-agent knowledge propagation and emergent coordination, contrasting with traditional passive vector databases.

Manticore's KNN search gets up to 29% faster with 2-pass HNSW, batched distances, compile-time distance specialization, and AVX-512 support.

Manticore Search introduces early termination for HNSW-based KNN vector search, reducing distance computations by up to 80% for large k values while maintaining precision within 2-4% of full search.