VBASE: Unifying Online Vector Similarity Search and Relational Queries via Relaxed...
summary
VBase identifies a property called relaxed monotonicity that lets approximate nearest-neighbor indices serve as termination checks for SQL-style queries, avoiding repeated TopK trials and proving equivalence to an optimal- K TopK result. Implemented on Postgres (open-sourced as msvbase), it accelerates vector+filter queries 2–10×, yields ~100× for multi-vector TopK and over 3 orders-of-magnitude speedup for vector joins, and reports up to 7,000× speedup with 99.9% accuracy on exact queries by unifying vector index traversal behavior with relational operators.
AI-written summary · about
chapters (7)
- 0:00Introduction to vectors
- 1:30Limitations of current search
- 3:43Challenges of SQL and indexing
- 5:50Defining relaxed monotonicity
- 9:28Designing the vbase system
- 10:54Evaluation and performance
- 12:22Conclusion and future work
related talks
48mFrom Text to Meaning: pgvector Transforms PostgreSQL Search
26mThe Rise of Vector Databases - Lessons from the Milvus Community
27mUnified AI & Relational Power: Scalable Distributed Vector Search with
33mBringing Vector Search to OLTP
15mUnconference: Explainable Vector Search
41mPostgreSQL for AI. Real Vector Search in PostgreSQL
44mVector Search Beyond the Hype: Real-World Implementation and Performance Insights
04mSupercharging GenAI PostgreSQL Vector Search Optimization