Vector Databases Are Broken?! ACORN’s Hybrid Search Breakthrough [SIGMOD 2024] with Liana Patel скачать в хорошем качестве

Vector Databases Are Broken?! ACORN’s Hybrid Search Breakthrough [SIGMOD 2024] with Liana Patel

Description Vector search falls apart the moment you add filters… or does it? 🎯 In this episode of Disseminate, host Jack Waudby sits down with Liana Patel (Stanford) to unpack ACORN, a predicate-agnostic, high-performance hybrid search index built on HNSW and presented at SIGMOD. We dive into why pre/post-filtering fail, how ACORN traverses predicate subgraphs, and what this means for RAG, recommendation, and real-world vector databases. What you’ll learn Why standard ANN + filters tanks recall/QPS—and how ACORN fixes it Predicate-agnostic hybrid search (structured + unstructured) without special-case hacks HNSW recap → ACORN’s dense graph + pruning (neighbors-of-neighbors) intuition ACORN-γ vs ACORN-1: build-time vs query-time trade-offs Benchmarks, query correlation, and where this goes next (serverless, updates) Chapters 00:00 Intro & guest 01:10 Liana’s path into ML systems & databases 03:30 Vector embeddings, ANN/HNSW 101 06:00 Why “hybrid search” is hard (filters + vectors) 12:00 ACORN: core idea & search over predicate subgraphs 20:40 Build vs search: ACORNγ vs ACORN-1 26:00 Benchmarks, recall/QPS, query correlation 35:10 Limitations, indexing time, future work 49:30 Vector DBs vs relational: where this is headed 51:20 One takeaway for practitioners Who’s this for? Industry practitioners shipping RAG/search, researchers in IR/DB/ML systems, and students wanting a crisp bridge from papers to practice. Links ACORN: Performant and Predicate-Agnostic Search Over Vector Embeddings and Structured Data [SIGMOD'24] Paper: https://arxiv.org/pdf/2403.04871 Liana's LinkedIn:   / liana-patel-b0a51316a   Liana's X: https://x.com/lianapatel_ #VectorDatabases #HybridSearch #ACORN #HNSW #ANN #VectorSearch #RAG #InformationRetrieval #SIGMOD #MLSystems #Databases #LLM #MachineLearning #SystemsResearch #Stanford