Vector Search with Elastic: Redefining Similarity: SF Elastic User Group Meetup

Talk Abstract: Robert Statsinger - Principal Solution Architect @ Elastic

Similarity between elements in a dataset has traditionally been measured based on appearance - simple measures such as word counts and other lexical similarities have been the state of the practice. Vector Search goes beyond appearances and lets you define similarity based on meanings and deeper representations of content. Image recognition and comparisons, audio comparisons and recommendations, and relevance ranking based on Natural Language Processing (NLP) are just a few of the applications that Vector Search enables. The Elastic Platform equips you with the tools you need to create novel applications based on this approach.

Highlights

-Define indexes to hold vectorized data using Elastic’s dense_vector field type
-Use vector similarity measures such as l2_norm, dot_product, and cosine
-Perform efficient searches of the data using a Hierarchical Navigable Small World (HNSW) search algorithm

Agenda

-Introduction and opening remarks
-Working with Vectorized Data
-Similarity for Vectors
-Search algorithms: KNN and ANN
-Demo
-Q&A

Here are the links for the hands-on labwork if you're interested: