We like to discuss the usability of new architecture of partitioned third party, offered in [1] for conducting a new protocols for data mining algorithms over shared data base between multiple data holders. Current solution for data mining over partitioned data base are: Data anonimization [4], homomorphic encryption [5], trusted third party [2] or secure multiparty computation algorithms [3]. Current solutions suffer from different problems such as expensive algorithms in terms of computation overhead and required communication rounds, revealing private information to third party. The new architecture offered by Sherman et al. allow the data holders to use simple masking techniques that are not expensive in computation nor assume trust in the third party, yet allow to perform simple and complex data mining algorithms between multiple data owners while private data is not revealed. That come with the assumption of no collude between the two parts of the PTTP.

1: Q sends the query to DBi, 1 ≤ i ≤ M, and the query type (either intersection or
union) to R.
2: Q sends R which private column participate in the query (e.g. the age column).
3: R generates a random permutation σ on the set of integers {1,…, 2|Ω|} and sends
it to DBi, 1 ≤ i ≤ M.
4: for all 1 ≤ i ≤ M do
5: DBi sets a Boolean vector Vi={vi,1,. . . ,vi,|Ω|} where
vi,j =
1 ifDBi hold private value j
0 otherwise