Vector data in Postgres – how’s it different from “normal” data?

Estimated read time 2 min read

Post Content

​ Video of a conference talk about vector data in Postgres presented by Heikki Linnakangas at POSETTE: An Event for Postgres 2024. AI works with vectors. But what is a vector? How is a vector in Postgres different from integers, text, geospatial data, or other traditional Postgres datatypes? Why can’t you create a regular B-tree index over vectors? Ok, but why not a GiST index?

Heikki will touch on these Postgres and pgvector topics:

– pgvector extension’s data types, including the new sparse and bit vector types
– Curse of dimensionality with vector databases
– Vector search algorithms, including HNSW (hierarchical navigable small worlds)

Heikki is a cofounder of Neon and a long-time PostgreSQL developer and committer. Since 2023, he has also been contributing to pgvector.

Chapters:
00:00 – Intro
01:37 – Vector search is expensive
02:30 – What is a vector?
03:57 – Vectors are large
06:36 – What is vector search?
11:45 – Vector search is approximate
15:17 – Types of vector search algorithms
20:12 – Three new datatypes in pgvector 0.7.0
23:55 – Summary

? Everything you need to know about POSETTE: An Event for Postgres 2024 can be found at: https://aka.ms/posette

✅ Learn more:
Watch more POSETTE talks: https://aka.ms/posette-playlist

? Let’s connect:
X – @PosetteConf https://x.com/PosetteConf
Mastodon – @posetteconf https://mastodon.social/@posetteconf
Threads – @posetteconf https://www.threads.net/@posetteconf

? Subscribe to the POSETTE News: https://aka.ms/posette-subscribe

? Slides for this talk: https://speakerdeck.com/the4thdoctor/vindicating-zfs-with-postgresql-unleashing-the-power-of-scalability-9a17f9c3-d69f-48f3-8a9f-c22e31c9ff0d

#PosetteConf #PostgreSQL #AI

Presenters:
Heikki Linnakangas, PostgreSQL hacker & committer, co-founder of Neon   Read More Microsoft Developer 

You May Also Like

More From Author

+ There are no comments

Add yours