# US 12,008,443 B2 · Method and apparatus with ciphertext training analysis model

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,008,443 B2
- Document ID: US12008443B2
- Status: Granted
- Kind: B2
- Application number: 17/081,519
- Filed: 2020-10-27
- Published: 2021-10-21
- Granted: 2024-06-11
- Earliest claimed: 2020-04-21
- Assignees: Samsung Sds Co., Ltd.
- Applicants: Samsung Sds Co., Ltd.
- Inventors: Eun Kyung Kim; Hyo Jin Yoon; Ji Hoon Cho; Min Jung Kim; Young Seon Lee
- Web page: https://patentgrep.com/patent/US12008443B2

## CPC

- G06N 20/00: Machine learning
- H04L 9/008: {involving homomorphic encryption}
- H04L 9/0618: {Block ciphers, i.e. encrypting groups of characters of a plain text message using fixed encryption transformation}
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)

## Abstract

A method for training an analysis model according to an embodiment includes receiving synthetic data for original data from a client apparatus, determining a hyperparameter value for analysis model training using the synthetic data, receiving a ciphertext for the original data from the client apparatus, and training the analysis model using the determined hyperparameter value and the received cipher text.

## Claims

### Claim 1

A processor-implemented method for training an analysis model, the method comprising:
receiving synthetic data for original data from a client apparatus;
determining a data preprocessing scheme to be applied to the original data for analysis model training using the synthetic data;
determining a hyperparameter value for training analysis model using the synthetic data;
receiving a ciphertext for the original data from the client apparatus; and
training the analysis model using the determined hyperparameter value and the received ciphertext.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
notifying the determined data preprocessing scheme to the client apparatus,
wherein the receiving of the ciphertext comprises receiving a ciphertext for data generated by preprocessing the original data according to the determined data preprocessing scheme.

### Claim 3 (depends on 1)

The method of claim 1, wherein the synthetic data satisfies local differential privacy.

### Claim 4 (depends on 1)

The method of claim 1, wherein the ciphertext is a ciphertext encrypted using a homomorphic encryption algorithm.

### Claim 5 (depends on 4)

The method of claim 4, wherein the training of the analysis model comprises training the analysis model by using the received ciphertext in an encrypted state.

### Claim 6

An apparatus for training an analysis model, comprising:
one or more processors configured to execute instructions; and
a memory storing the instructions, wherein execution of the instructions configures the processors to:
receive synthetic data for original data from a client apparatus;
determine, based on the synthetic data, a data preprocessing scheme to be applied to the original data for analysis model training;
determine a hyperparameter value using the synthetic data;
receive a ciphertext for the original data from the client apparatus; and
train the analysis model using the determined hyperparameter value and the received ciphertext.

### Claim 7 (depends on 6)

The apparatus of claim 6, wherein the receiving the synthetic data further comprises
notifying the determined data preprocessing scheme to the client apparatus, and
wherein the receiving the ciphertext further comprises receiving a ciphertext for data generated by preprocessing the original data according to the determined data preprocessing scheme.

### Claim 8 (depends on 6)

The apparatus of claim 6, wherein the synthetic data satisfies local differential privacy.

### Claim 9 (depends on 6)

The apparatus of claim 6, wherein the ciphertext is a ciphertext encrypted using a homomorphic encryption algorithm.

### Claim 10 (depends on 9)

The apparatus of claim 9, wherein the receiving the ciphertext further comprises training the analysis model by using the received ciphertext in an encrypted state.

### Claim 11

A method performed by a computing device in communication with a client device, the method comprising:
receiving, from the client device, synthetic data having a same statistical characteristic as original data stored on the client device;
based on receiving the synthetic data, determining a hyperparameter value of a neural network model that optimizes training of the neural network model;
receiving from the client device a ciphertext generated by encryption of the original data;
configuring the neural network model according to the hyperparameter value and training the neural network model based on the ciphertext.

### Claim 12 (depends on 11)

The method of claim 11, further comprising:
based on the received synthetic data, selecting, from among a set of predefined preprocessing schemes, a preprocessing scheme determined to be optimal for training the neural network model.

### Claim 13 (depends on 12)

The method of claim 12, further comprising:
sending an indication of the selected preprocessing scheme to the client device, and wherein
based on the indication, the client device uses the preprocessing scheme to generate the ciphertext before sending the ciphertext to the computing device.

### Claim 14 (depends on 12)

The method of claim 12, wherein the determining the hyperparameter value comprises iteratively training the neural network model based on the synthetic data.

### Claim 15 (depends on 12)

The method of claim 12, wherein the preprocessing scheme is selected by performing training of the neural network model based on the synthetic data.