US 12,483,381 B2
Non-linear approximation robust to input range of homomorphic encryption analytics
Omri Soceanu, Haifa (IL); Allon Adir, Kiryat Tivon (IL); Omer Yehuda Boehm, Haifa (IL); Boris Rozenberg, Ramat Gan (IL); Eyal Kushnir, Kfar Vradim (IL); and Ehud Aharoni, Kfar Saba (IL)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Jun. 8, 2023, as Appl. No. 18/207,187.
Prior Publication US 2024/0413966 A1, Dec. 12, 2024
Int. Cl. H04L 9/00 (2022.01); G06N 5/04 (2023.01)
CPC H04L 9/008 (2013.01) [G06N 5/04 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A method for privacy-preserving homomorphic inferencing, comprising:
receiving an encrypted data point at a machine learning model, the machine learning model having an activation function, the encrypted data point comprising an input feature vector that has been extended with a set of one or more additional feature values, the set of one or more additional feature values having been generated by applying a normalized inverse function to respective one or more features in the input feature vector, wherein the normalized inverse function computes a reciprocal value of each feature value in the input feature vector and normalizes resulting reciprocal values into a common range, and wherein the set of one or more additional feature values comprises a reciprocal value for every input feature value in the input feature vector;
performing homomorphic inferencing on the encrypted data point using the machine learning model to generate an encrypted result; and
returning the encrypted result in response to the encrypted data point.