The article surveys emerging vector databases, explains how various vector‑search algorithms such as FLAT, IVF, HNSW, DiskANN and ScaNN differ in accuracy, speed, memory use and build time, and provides practical guidance for choosing the right index based on data size, latency and resource constraints.
This article explains how unstructured multimedia data is transformed into vectors for similarity search, introduces GPU parallelism and the NN‑Descent algorithm to replace traditional HNSW indexing in OpenSearch, and presents benchmark results showing up to a thirteen‑fold speed improvement while maintaining comparable recall.
At the ACM CCS 2025 live paper showcase, three groundbreaking studies—Gibbon’s fast secure two‑party GBDT training, Panther’s efficient private approximate nearest‑neighbor search on a single server, and PromeFuzz’s knowledge‑driven LLM approach to fuzzing harness generation—are presented, highlighting significant performance and security advances.
This article explains the principles and practical implementation of vector retrieval in StarRocks, covering approximate nearest‑neighbor algorithms, index design, query planning, performance optimizations, real‑world case studies, and future challenges, showing how query latency dropped from 15 seconds to 2 seconds while cutting costs to a third.
This article explains how OpenSearch's GPU‑accelerated vector search leverages parallel graph algorithms to achieve up to tenfold speed improvements over CPU solutions, detailing ANNS techniques, performance benchmarks, and practical GPU specifications for high‑QPS AI applications.
RTAMS‑GANNS, the award‑winning real‑time adaptive multi‑stream GPU system for online approximate nearest neighbor search, eliminates costly memory allocations and serial execution by using a dynamic memory‑block insertion algorithm and separate CUDA streams, cutting latency by 40‑80% and reliably serving over 100 million daily users in production.
Vector retrieval, based on converting data into high‑dimensional vectors and measuring similarity, enables fast, accurate search across massive datasets, supporting AI tasks such as search engines, recommendation, NLP, and computer vision, and plays a crucial role in large‑model training for data selection, anomaly detection, and model optimization.
Baidu’s upcoming technical salon series on search will showcase its open‑source ANN retrieval framework Puck—demonstrating superior performance on massive benchmarks—and feature nightly live sessions covering product management, GPU optimization, intelligent Q&A, model architecture, and image generation, with registration via QR‑code.
The article explains how vectors represent data in high‑dimensional space, describes the embedding process, outlines the evolution and challenges of vector search, compares exact and approximate algorithms such as IVF, product quantization and HNSW, and details Baidu’s cloud‑native engineering solutions for scalable, filtered vector retrieval.
The article explains how combining vector databases with large language models can help governments and enterprises leverage massive private data for AI, covering semantic search, approximate nearest neighbor techniques, alignment challenges across modalities, and future directions for fine‑grained data integration.
This paper identifies the trade‑off between simple and hard negatives in embedding‑based retrieval for recommendation, proposes a clustering‑based divide‑and‑conquer framework combined with prompt‑driven multi‑task learning to improve relevance, diversity, and fairness, and validates the approach through offline metrics, online A/B tests, and comparative experiments.
This tutorial explains what Faiss is, how to install it, construct various indexes such as IndexFlatL2, IndexIVFFlat, and IndexIVFPQ, and demonstrates code examples for building and querying vector similarity search pipelines while discussing speed‑accuracy trade‑offs.
The inaugural BigANN competition, organized by NeurIPS, showcased large‑scale ANN research, and Baidu's self‑developed PUCK algorithm secured top scores across all four tracks by leveraging multi‑layer quantization, two‑level inverted indexing, and extensive system‑level optimizations.
This article introduces the concept of vector retrieval, outlines its diverse application scenarios, explains common distance metrics for both floating‑point and binary vectors, and surveys fundamental approximate nearest‑neighbor algorithms including tree‑based, graph‑based, quantization, and hashing methods.
This article explains how deep learning transforms raw physical‑world data into dense vectors, defines the significance of vector retrieval, surveys common use cases such as image, video, and text search, discusses challenges in representation learning, and reviews high‑performance approximate nearest‑neighbor algorithms and practical deployments.
This article explains how deep learning transforms unstructured data into dense vectors, defines vector retrieval, outlines its many use cases such as product, video, and text search, discusses challenges in learning effective embeddings, and reviews high‑performance algorithms like LSH, neighbor graphs, and product quantization.
The Hengyang Data team won the 2019 CCF Big Data & Computing Intelligence video‑copyright detection contest by extracting VGG16‑based image features with Gaussian‑R‑MAC weighting, using a graph‑based NSG nearest‑neighbor search and a frame‑matching algorithm to locate infringing segments within three‑second precision, even under severe cropping and other transformations.
