Granted Patent B2

US 12,425,181 · App. 17/853,798 · Granted Sep 23, 2025

Inventors: Jae Wook Lee (Seoul, KR); Jun Young Byun (Seoul, KR)

Assignee: SEOUL NATIONAL UNIVERSITY R&DB FOUNDATION

H04L9/008G06F18/2415

Abstract

Claims (44)

1. A homomorphic encryption data processing method comprising:

deriving a first classification value which is a segment classification value of a dataset including homomorphic encryption by inputting the dataset of data including the homomorphic encryption to a segment classification module for determining whether the data is about what kind of information, wherein the segment classification module was trained by using a non-homomorphic encryption dataset that does not include the homomorphic encryption;

deriving a first distribution average value of the first classification value and deriving a second distribution average value by adding a weighted value to the first distribution average value;

deriving a second classification value by using the second distribution average value; and

performing computation of data including the homomorphic encryption by using the second classification value and the second distribution average value,

wherein the weighted value is used to correct a distribution bias that occurs when data including homomorphic encryption is input to the segment classification module.

2. The homomorphic encryption data processing method of claim 1 , wherein the deriving of the first classification value comprises:

extracting a first logit of the dataset including the homomorphic encryption; and

setting, as the first classification value, a value corresponding to probability that the data of the dataset including the homomorphic encryption from the first logit is classified into any one of segments.

3. The homomorphic encryption data processing method of claim 2 , wherein the deriving of the second distribution average value comprises:

deriving a third classification value that is a segment classification value of the dataset that does not include the homomorphic encryption by inputting the dataset of the data that do not include the homomorphic encryption to the segment classification module;

deriving a third distribution average value that is a distribution average value of the third classification value; and

setting the weighted value such that the first distribution average value is equal to the third distribution average value by performing average fitting of the first distribution average value and the third distribution average value.

4. The homomorphic encryption data processing method of claim 3 , wherein the performing of the computation of the data including the homomorphic encryption comprises:

storing a first dataset including data including the homomorphic encryption and data that do not include the homomorphic encryption as a slot;

calculating an estimation value for a previously stored linear model by applying the homomorphic encryption data and the non-homomorphic encryption data in the first dataset to the linear model; and

transmitting the calculated estimation value to an external apparatus.

5. The homomorphic encryption data processing method of claim 4 , wherein the calculating of the estimation value comprises extracting a term using the homomorphic encryption data from among polynomials required for calculating the estimation value of the linear model and calculating the estimation value by performing homomorphic computation for the extracted term.

6. The homomorphic encryption data processing method of claim 5 , wherein the calculating of the estimation value comprises:

generating a matrix corresponding to the first dataset;

decomposing the matrix into a first matrix including the homomorphic encryption data and a second matrix that does not include the homomorphic encryption data;

enabling the second matrix to be orthogonal to the first matrix; and

calculating the estimation value for the linear model by using the first matrix and the second matrix.

7. The homomorphic encryption data processing method of claim 6 , wherein the first dataset comprises a plurality of different variables, each homomorphically encrypted.

8. The homomorphic encryption data processing method of claim 7 , wherein the linear model is a ridge regression linear model.

9. A homomorphic encryption data computing apparatus comprising:

a memory storing at least one instruction; and

a processor configured to execute the at least one instruction,

wherein the processor executes the at least one instruction to:

classify whether data is about what kind of information,

derive a first classification value that is a segment classification value of a dataset of data including a homomorphic encryption by inputting the dataset of data including the homomorphic encryption to a segment classification module, wherein the segment classification module was trained by using a non-homomorphic encryption dataset that does not include the homomorphic encryption,

derive a first distribution average value of the first classification value,

drive a second distribution average value by adding a weighted value to the first distribution average value,

drive a second classification value by using the second distribution average value, and

calculate data including a homomorphic encryption by using the second classification value and the second distribution average value,

wherein the weighted value is used to correct a distribution bias that occurs when data including homomorphic encryption is input to the segment classification module.

10. The homomorphic encryption data computing apparatus of claim 9 , wherein the processor extracts a first logit of the dataset including the homomorphic encryption, and sets, as the first classification value, a value corresponding to probability that the data of the dataset including the homomorphic encryption from the first logit is classified into any one of segments.

11. The homomorphic encryption data computing apparatus of claim 10 , wherein the processor derives a third classification value which is a segment classification value of a dataset that does not include the homomorphic encryption by inputting the dataset of the data that do not include the homomorphic encryption to the segment classification module, derives a third distribution average value that is a distribution average value of the third classification value, and sets the weighted value such that the first distribution average value is equal to the third distribution average value by performing average fitting of the first distribution average value and the third distribution average value.

12. The homomorphic encryption data computing apparatus of claim 11 , wherein the processor performs repetitive learning of a classification process as to whether data is homomorphic encryption data by using the non-homomorphic encryption dataset.

13. The homomorphic encryption data computing apparatus of claim 12 , wherein the processor stores a first dataset including data including the homomorphic encryption and data that do not include the homomorphic encryption as a slot and calculates an estimation value for a previously stored linear model by applying the homomorphic encryption data and the non-homomorphic encryption data in the first dataset to the linear model.

14. The homomorphic encryption data computing apparatus of claim 13 , wherein the processor extracts a term using the homomorphic encryption data from among polynomials required for calculating the estimation value of the linear model and calculating the estimation value by performing homomorphic computation for the extracted term.

15. The homomorphic encryption data computing apparatus of claim 14 , wherein the processor generates a matrix corresponding to the first dataset, decomposes the matrix into a first matrix including the homomorphic encryption data and a second matrix that does not include the homomorphic encryption data, enables the second matrix to be orthogonal to the first matrix, and calculates the estimation value for the linear model by using the first matrix and the second matrix.

16. The homomorphic encryption data computing apparatus of claim 15 , wherein the first dataset comprises a plurality of different variables, each homomorphically encrypted.

17. The homomorphic encryption data computing apparatus of claim 16 , wherein the linear model is a ridge regression linear model.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (11)