top of page

AI Oriented Homomorphic Encryption

How to Use Untrusted AI to detect patterns in your secret data

The early use of homorphic encryption was focused on pre defined mathematical operations that can be carried out over encrypted data so that upon decryption they yield the same result as if the operation was carried out on the pre-encrypted data. This hides the input data from the party that carries out the computation. Obviously, the application range was quite limited, but the cryptographic body of solutions was intellectually rewarding.

A different kind of homomorphic encryption is now emerging as a need - with a broad range of applications. At issue is the common reality where AI engines are chewing large amounts of data searching for patterns, and identifying inferential pathways. While most of the data feed for AI engines is public data, it is of great need for private organizations to let AI go over secret data to extract useful patterns. If the party that applies the AI is not fully trusted then homomorphic encryption is called for. Alas, here the encryption cannot be limited to a pre specified mathematical operation, it has to be wide open

That is the challenge that is met by BitMint Homomorphic Encryption tools. We use two approaches:

               1. Numerization

               2. Similarity Calculus

Numerization (see white paper) transforms the feed data into a fuzzy version where patterns survive, but specificity does not. Similarity calculus replaces the original data with similarity indices, which are more readily digested by the various AI inferential engines (see blog).

homomorphic.png
Address

 

BitMint

P.O.Box 1022, McLean, VA 22101, USA



info@BitMint.com

 

Inquire about BitMint Homomorphic Encryption

Your details were sent successfully!

bottom of page