Fig. 3: Multiple imputation via MHE.
From: Secure distributed multiple imputation enables missing data inference for private data proprietors
The input data is distributed between the parties and kept in a non-encrypted form, only to be encrypted when needed during the imputation and final analysis. The procedure also benefits from independent, parallel computation on top of local data partitions. This scheme, however, is suitable only for large-scale datasets due to the performance overhead incurred by the underlying, expensive cryptographic scheme that is inherently scalable with respect to data size and the number of computing parties.
