Fast and secure similarity search in high dimensional space

Similarity search in high dimensional space database is split into two worlds: i) fast, scalable, and approximate search algorithms which are not secure, and ii) search protocols based on secure computation which are not scalable. This paper presents a one-way privacy protocol that lies in between these two worlds. Approximate metrics for the cosine similarity allows speed. Elements of large random matrix theory provides security evidences if the size of the database is not too big with respect to the space dimension.