Privacy-Preserving, Efficient, and Accurate Dimensionality Reduction

Summary

Researchers propose PP-DR, a privacy-preserving dimensionality reduction scheme that secures participant data with homomorphic encryption. It avoids bootstrapping by using secure interaction protocols, so participants can jointly reduce shared data. The authors report an average error of only 1*10^-8, a 30 to 200 times gain in computational efficiency, and at least a 70% reduction in communication overhead compared with prior homomorphic-encryption dimension reduction methods.