Harnessing the Universal Geometry of Embeddings

TL;DR

This paper presents an unsupervised method for translating text embeddings between different vector spaces without paired data, enabling cross-model compatibility and raising security concerns for embedding privacy.

Contribution

It introduces the first universal, unsupervised translation method for embeddings that does not require paired data or encoders, based on a universal semantic structure hypothesis.

Findings

High cosine similarity achieved across diverse models

Effective translation without paired data

Implications for embedding security and privacy

Abstract

We introduce the first method for translating text embeddings from one vector space to another without any paired data, encoders, or predefined sets of matches. Our unsupervised approach translates any embedding to and from a universal latent representation (i.e., a universal semantic structure conjectured by the Platonic Representation Hypothesis). Our translations achieve high cosine similarity across model pairs with different architectures, parameter counts, and training datasets. The ability to translate unknown embeddings into a different space while preserving their geometry has serious implications for the security of vector databases. An adversary with access only to embedding vectors can extract sensitive information about the underlying documents, sufficient for classification and attribute inference.