US-12619685-B2 - Apparatus for classifying data and method thereof

US12619685B2US 12619685 B2US12619685 B2US 12619685B2US-12619685-B2

Abstract

Provided is a method for classifying data in an electronic apparatus, including obtaining target data, obtaining first classification information by using a classifier set including a plurality of classifiers based on the target data, obtaining second classification information by using a neural network model based on the target data, comparing the first classification information and the second classification information, and verifying the classifier set based on a result of comparing the first classification information and the second classification information.

Inventors

Yihwan Kim
Hoseok Chang
Namsoon Jung

Assignees

AiM Future Inc.

Dates

Publication Date: 20260505
Application Date: 20230306

Claims (20)

1 . A method for classifying data in an electronic apparatus, comprising: obtaining target data; obtaining first classification information by using a classifier set including a plurality of classifiers based on the target data; obtaining second classification information by using a neural network model based on the target data; comparing the first classification information and the second classification information; and verifying the classifier set based on a result of comparing the first classification information and the second classification information, wherein at least one of the first classification information and the second classification information includes a plurality of candidate classification values, wherein if the second classification information includes the plurality of candidate classification values, the comparing the first classification information and the second classification information includes: determining one or more candidate classification values that satisfy a configured criterion among the plurality of candidate classification values included in the second classification information; and identifying whether each of the one or more candidate classification values matches a classification value indicated by the first classification information, wherein the configured criterion is a criterion related to probabilities that data belongs to each of the plurality of candidate classification values, and determining by the electronic apparatus, scheduling of an operation performed by a core, among a plurality of cores of a processor, to be assigned to the operation for each of the plurality of classifiers based on each classifier requirements for an amount of computation needed for the operation and computing power of each of the plurality of cores.
2 . The method of claim 1 , further comprising, based on the result of comparing the first classification information and the second classification information, determining an operating method of the classifier set, wherein the operating method of the classifier set includes an operation sequence of the plurality of classifiers.
3 . The method of claim 2 , wherein the electronic apparatus determines the operating method of the classifier set further based on information of time required for the classifier set to generate the first classification information.
4 . The method of claim 2 , wherein the determining the operating method of the classifier set, includes: based on the result of comparing the first classification information and the second classification information, determining reliability of the first classification information; determining whether the reliability exceeds a configured threshold value; and if the reliability is less than the configured threshold value, changing the operating method of the classifier set.
5 . The method of claim 1 , wherein the obtaining the first classification information includes: using a first classifier included in the plurality of classifiers based on the target data; if output information of the first classifier satisfies a configured condition, determining the output information of the first classifier as the first classification information; and if the output information of the first classifier does not satisfy the configured condition, if a second classifier configured to operate in an order following the first classifier is present among the plurality of classifiers, using the second classifier based on the target data; and if the second classifier is not present, obtaining information indicating that the configured condition is not satisfied.
6 . The method of claim 1 , wherein the first classification information and the second classification information include at least one of information on whether the target data is abnormal data, and if the target data is abnormal, information on a reason for an abnormality.
7 . The method of claim 1 , further comprising accessing a processor for neural processing, wherein the processor for neural processing processes at least some of an operation for obtaining the first classification information, an operation for obtaining the second classification information and an operation for comparing the first classification information and the second classification information.
8 . The method of claim 1 , wherein a first classifier included in the plurality of classifiers determines whether data has a first classification value, and wherein a second classifier included in the plurality of classifiers determines whether data has a second classification value, wherein the first classification value and the second classification value are different from each other.
9 . The method of claim 1 , wherein the obtaining the target data includes: obtaining original data; and generating the target data by processing the original data.
10 . The method of claim 1 , wherein the neural network model includes one or more among one or more convolution layers, one or more pooling layers and one or more dense layers.
11 . A computer-readable non-transitory recording medium having a program for executing a method on a computer, the method comprising: obtaining target data; obtaining first classification information by using a classifier set including a plurality of classifiers based on the target data; obtaining second classification information by using a neural network model based on the target data; comparing the first classification information and the second classification information; and verifying the classifier set based on a result of comparing the first classification information and the second classification information, wherein at least one of the first classification information and the second classification information includes a plurality of candidate classification values, wherein, if the second classification information includes the plurality of candidate classification values, the comparing the first classification information and the second classification information includes: determining one or more candidate classification values that satisfy a configured criterion among the plurality of candidate classification values included in the second classification information; and identifying whether each of the one or more candidate classification values matches a classification value indicated by the first classification information, wherein the configured criterion is a criterion related to probabilities that data belongs to each of the plurality of candidate classification values, and determining by the computer, scheduling of an operation performed by a core, among a plurality of cores of a processor, to be assigned to the operation for each of the plurality of classifiers based on each classifier requirements for an amount of computation needed for the operation and computing power of each of the plurality of cores.
12 . A method for classifying data in an electronic apparatus, comprising: obtaining target data; obtaining first classification information by using a classifier set including a plurality of classifiers based on the target data; if the first classification information satisfies a configured condition, determining a classification value of the target data based on the first classification information; and if the first classification information does not satisfy the configured condition, obtaining second classification information by using a neural network model based on the target data; comparing the first classification information and the second classification information; and determining a classification value of the target data based on the second classification information, wherein at least one of the first classification information and the second classification information includes a plurality of candidate classification values, wherein, if the second classification information includes the plurality of candidate classification values, the comparing the first classification information and the second classification information includes: determining one or more candidate classification values that satisfy a configured criterion among the plurality of candidate classification values included in the second classification information; and identifying whether each of the one or more candidate classification values matches a classification value indicated by the first classification information, wherein the configured criterion is a criterion related to probabilities that data belongs to each of the plurality of candidate classification values, and determining by the electronic apparatus, scheduling of an operation performed by a core, among a plurality of cores of a processor, to be assigned to the operation for each of the plurality of classifiers based on each classifier requirements for an amount of computation needed for the operation and computing power of each of the plurality of cores.
13 . An electronic apparatus for classifying data, comprising: a memory for storing instructions; and a processor, wherein the processor, being connected to the memory, is configured to: obtain target data; obtain first classification information by using a classifier set including a plurality of classifiers based on the target data; obtain second classification information by using a neural network model based on the target data; compare the first classification information and the second classification information; and verify the classifier set based on a result of comparing the first classification wherein at least one of the first classification information and the second classification information includes a plurality of candidate classification values, wherein, if the second classification information includes the plurality of candidate classification values, the comparing the first classification information and the second classification information includes: determining one or more candidate classification values that satisfy a configured criterion among the plurality of candidate classification values included in the second classification information; and identifying whether each of the one or more candidate classification values matches a classification value indicated by the first classification information, wherein the configured criterion is a criterion related to probabilities that data belongs to each of the plurality of candidate classification values, and determine by the electronic apparatus, scheduling of an operation performed by a core, among a plurality of cores of the processor, to be assigned to the operation for each of the plurality of classifiers based on each classifier requirements for an amount of computation needed for the operation and computing power of each of the plurality of cores.
14 . The electronic apparatus of claim 13 , further comprising, based on the result of comparing the first classification information and the second classification information, determining an operating method of the classifier set, wherein the operating method of the classifier set includes an operation sequence of the plurality of classifiers.
15 . The electronic apparatus of claim 14 , wherein the electronic apparatus determines the operating method of the classifier set further based on information of time required for the classifier set to generate the first classification information.
16 . The electronic apparatus of claim 14 , wherein the determining the operating method of the classifier set, includes: based on the result of comparing the first classification information and the second classification information, determining reliability of the first classification information; determining whether the reliability exceeds a configured threshold value; and if the reliability is less than the configured threshold value, changing the operating method of the classifier set.
17 . The electronic apparatus of claim 13 , wherein the obtaining the first classification information includes: using a first classifier included in the plurality of classifiers based on the target data; if output information of the first classifier satisfies a configured condition, determining the output information of the first classifier as the first classification information; and if the output information of the first classifier does not satisfy the configured condition, if a second classifier configured to operate in an order following the first classifier is present among the plurality of classifiers, using the second classifier based on the target data; and if the second classifier is not present, obtaining information indicating that the configured condition is not satisfied.
18 . The electronic apparatus of claim 13 , wherein the first classification information and the second classification information include at least one of information on whether the target data is abnormal data, and if the target data is abnormal, information on a reason for an abnormality.
19 . The electronic apparatus of claim 13 , further comprising accessing a processor for neural processing, wherein the processor for neural processing processes at least some of an operation for obtaining the first classification information, an operation for obtaining the second classification information and an operation for comparing the first classification information and the second classification information.
20 . The electronic apparatus of claim 13 , wherein a first classifier included in the plurality of classifiers determines whether data has a first classification value, wherein a second classifier included in the plurality of classifiers determines whether data has a second classification value, and wherein the first classification value and the second classification value are different from each other.

Description

CROSS-REFERENCE TO RELATED APPLICATION(S) This application claims the benefit of U.S. Provisional Patent Application No. 63/326,133, filed on Mar. 31, 2022, in the United States Patent and Trademark Office, the disclosure of which is incorporated herein by reference. BACKGROUND Technical Field Example embodiments relate to a method for obtaining first classification information using a classifier set including a plurality of classifiers, obtaining second classification information using a neural network model, and verifying the classifier set by comparing the first classification information and the second classification information, and relate to an apparatus thereof. Further, example embodiments relate to a method of obtaining first classification information using a classifier set, determining a classification value of target data based on the first classification information when the first classification information satisfies a configured condition, and if the first classification information does not satisfy the configured condition, obtaining second classification information using a neural network model to determine a classification value of the target data. Description of the Related Art The process of image recognition is frequently attempted in the fields of AI. It is also one of the most important applications in the relevant industry. Recognizing people or objects in the input images and utilizing the result of the recognition are useful in many of the practical applications in edge devices, such as the image recognition and processing through a smartphone. This kind of application is becoming more feasible because the performance of the built-in application processor(s) and camera(s) in smartphones has greatly improved in recent years. A neural processing unit (NPU) or AI accelerator can be used along with the powerful central processing unit (CPU) in these type of edge devices in order to provide more efficient AI functionalities, such as inferencing for the image recognition. In such cases, an optimization of the NPU is still important in the edge devices. It is because there is a limit to raising the performance of hardware itself for increasing the inference speed in an efficient manner. Further, a method for data classification using a neural network model may show high classification accuracy when sufficiently learned, but there are disadvantages such as requiring a large amount of computation and requiring secured sufficient training data. Therefore, if sufficient computing power is not secured, it is impossible to proceed with data classification using a neural network model, or even if it is possible to proceed with data classification using a neural network model, it may not be effective in terms of cost, such as taking an excessively long time for classification. With regard thereto, prior arts KR102407730B1 and US20210365774A1 may be referred to. BRIEF SUMMARY One or more embodiments of the present disclosure relates to an approach for detecting abnormal objects utilizing a NPU, and a novel approach for a hierarchical and optimized architecture of the multiple neural networks for the abnormal object detection. An aspect provides a method for classifying data in an electronic apparatus, including obtaining target data, obtaining first classification information by using a classifier set including a plurality of classifiers based on the target data, obtaining second classification information by using a neural network model based on the target data, comparing the first classification information and the second classification information, and verifying the classifier set based on a result of comparing the first classification information and the second classification information, and provides an electronic apparatus thereof. Another aspect also provides a method for classifying data in an electronic apparatus, including obtaining target data, obtaining first classification information by using a classifier set including a plurality of classifiers based on the target data, if the first classification information satisfies a configured condition, determining a classification value of the target data based on the first classification information, and if the first classification information does not satisfy the configured condition, obtaining second classification information by using a neural network model based on the target data and determining a classification value of the target data based on the second classification information. The technical tasks to be achieved by the present example embodiments are not limited to the technical tasks described above, and other technical tasks may be inferred from the following example embodiments. According to an aspect, there is provided a method for classifying data in an electronic apparatus, including obtaining target data, obtaining first classification information by using a classifier set including a plurality of classifiers based on the ta