vec2vec-no-paired-data-no-encoder-access
IN premise — summaries/2026/08/24/jha-2025-vec2vec-s9-discussion-and-future-work.md
Created 2026-08-24T17:10:59+00:00
vec2vec requires zero paired input-output examples across embedding spaces and zero access to the source encoder that produced the input embeddings
Summary
vec2vec can shrink high-dimensional embeddings to fewer dimensions using only the numbers in the vectors themselves, without needing the original model that created them or any labeled correspondences between different embedding spaces. This makes it a practical, model-agnostic compression step you can apply to any set of vectors the moment you have them, with no extra data or access to require.