Granted Patent B2

US 12,321,842 · App. 17/505,259 · Granted Jun 3, 2025

Inventors: Jinsu Yeo (Suwon-si, KR); Youngyoon Lee (Suwon-si, KR); Youngcheon You (Suwon-si, KR); Jaechool Lee (Suwon-si, KR)

Assignee: SAMSUNG ELECTRONICS CO., LTD.

G06N3/04G06F11/3409H03M7/702G06N3/044G06N3/045G06N3/0464G06N3/0495G06N3/08G06N3/082G06N3/09G06N3/0985H03M7/3059H03M7/70

Abstract

Claims (58)

1. A method for controlling an electronic apparatus, the method comprising:

receiving target data to be input as an input value of an operation;

selecting a generic-purpose artificial intelligence model;

generating a compressed artificial intelligence model based on the selected generic-purpose artificial intelligence model, wherein the generating of the compressed artificial intelligence model comprises compressing the selected generic-purpose artificial intelligence model using a low rank approximation algorithm by applying a singular value decomposition (SVD) algorithm for one matrix, and training the compressed artificial intelligence model;

generating a dedicated artificial intelligence model based on the compressed artificial intelligence model;

acquiring output data by performing the operation using the dedicated artificial intelligence model; and

providing the output data through an output interface,

wherein the generating of the compressed artificial intelligence model further comprises:

acquiring a rank of the SVD algorithm based on a compression rate,

compressing and training the selected generic-purpose artificial intelligence model based on the acquired rank and converting the compressed artificial intelligence model into a converted compressed artificial intelligence model,

determining a performance of the converted compressed artificial intelligence model based on a predetermined first threshold value, and

based on the performance of the converted compressed artificial intelligence model being lower than the predetermined first threshold value, generating the dedicated artificial intelligence model.

2. The method of claim 1 , wherein the generating a of the compressed artificial intelligence model comprises:

based on the performance of the converted compressed artificial intelligence model being higher than or equal to the predetermined first threshold value, repeating the generating a compressed of the compressed artificial intelligence model.

3. The method of claim 1 , wherein the generating of the dedicated artificial intelligence model comprises:

retraining the compressed artificial intelligence model based on a predetermined learning data set.

4. The method of claim 3 , wherein the generating of the dedicated artificial intelligence model comprises:

determining a performance of the retrained compressed artificial intelligence model in relation to a predetermined second threshold value; and

based on the performance of the retrained compressed artificial intelligence model being higher than or equal to the predetermined second threshold value, additionally compressing the retrained compressed artificial intelligence model and generating the dedicated artificial intelligence model.

5. The method of claim 3 , wherein the generating of the dedicated artificial intelligence model comprises:

based on feature information of first learning data that trained the selected generic-purpose artificial intelligence model, classifying the first learning data into a plurality of learning data sets, and reconstituting at least one data set among the plurality of learning data sets as the predetermined learning data set.

6. The method of claim 5 , wherein the predetermined learning data set comprises at least one data set compressed to correspond to the target data input to be as the input value among the plurality of learning data sets.

7. The method of claim 3 , wherein the generating of the compressed artificial intelligence model comprises:

determining a performance of the converted compressed artificial intelligence model based on test data, and

wherein the generating of the dedicated artificial intelligence model comprises:

classifying the test data into a plurality of learning data sets, and reconstituting at least one data set among the plurality of learning data sets as the predetermined learning data set.

8. The method of claim 7 ,

wherein the test data includes test data that output a result within a predetermined range based on a target value.

9. The method of claim 3 , wherein the generating of the dedicated artificial intelligence model comprises:

based on receiving data setting information from a user, identifying a learning data set corresponding to the data setting information from a first learning data that trained the selected generic-purpose artificial intelligence model, and reconstituting the identified learning data set as the predetermined learning data set.

10. An electronic apparatus comprising:

a memory storing first learning data and a generic-purpose artificial intelligence model trained by the first learning data; and

a processor,

wherein the processor is configured to:

receive target data to be input as an input value of an operation,

select the trained generic-purpose artificial intelligence model,

acquire a rank of a singular value decomposition (SVD) algorithm based on a compression rate,

compress and train the selected generic-purpose artificial intelligence model based on the acquired rank and convert the compressed artificial intelligence model into a converted compressed artificial intelligence model,

determine a performance of the converted compressed artificial intelligence model based on a predetermined first threshold value,

based on the performance of the converted compressed artificial intelligence model being lower than the predetermined first threshold value, generate a dedicated artificial intelligence model,

acquire output data by performing the operation using the dedicated artificial intelligence model, and

provide the output data through an output interface,

wherein the processor is further configured to:

compress the selected generic-purpose artificial intelligence model using a low rank approximation algorithm by applying the SVD algorithm for one matrix, and

train the compressed artificial intelligence model.

11. The electronic apparatus of claim 10 , wherein the processor is further configured to:

based on the performance of the converted compressed artificial intelligence model being higher than or equal to the predetermined first threshold value, repeat the converting of the compressed artificial intelligence model into the converted compressed artificial intelligence model.

12. The electronic apparatus of claim 10 , wherein the processor is further configured to:

retrain the compressed artificial intelligence model based on a predetermined learning data set.

13. The electronic apparatus of claim 12 , wherein the processor is further configured to:

determine a performance of the retrained compressed artificial intelligence model in relation to a predetermined second threshold value; and

based on the performance of the retrained compressed artificial intelligence model being higher than or equal to the predetermined second threshold value, additionally compress the retrained compressed artificial intelligence model and generate the dedicated artificial intelligence model.

14. The electronic apparatus of claim 12 , wherein the processor is further configured to:

based on feature information of the first learning data that trained the selected generic-purpose artificial intelligence model, classify the first learning data into a plurality of learning data sets, and reconstitute at least one data set among the plurality of learning data sets as the predetermined learning data set.

15. The electronic apparatus of claim 12 , wherein the processor is further configured to:

determine the performance of the compressed artificial intelligence model based on test data;

classify the test data into a plurality of learning data sets; and

reconstitute at least one data set among the plurality of learning data sets as the predetermined learning data set.

Assignments (1)

Priority Claims (1)

Continuity (2)

References Cited (32)