KR-20260064467-A - Method, apparatus, system and computer program for hierarchical classification using Mobius transformation

KR20260064467AKR 20260064467 AKR20260064467 AKR 20260064467AKR-20260064467-A

Abstract

The present invention relates to an artificial intelligence-based hierarchical classification method, device, system, and computer program, and more specifically, to an artificial intelligence-based hierarchical classification method, device, system, and computer program capable of performing a simulation by reflecting the impact of events, etc. detected based on artificial intelligence on the supply chain, and further deriving an optimal supply chain management plan. More specifically, the present invention discloses a method for performing a simulation of a supply chain using a computing device, comprising: a step of generating a plurality of simulation scenarios according to an event for a supply chain model that reflects an event given to the supply chain; a step of deriving a plurality of simulation results for the plurality of simulation scenarios based on the supply chain model; and a step of calculating a management plan for the supply chain corresponding to the event based on the plurality of simulation results.

Inventors

김성윤
김영준
문기효

Assignees

삼성에스디에스 주식회사

Dates

Publication Date: 20260507
Application Date: 20250611
Priority Date: 20241031

Claims (19)

A method for performing classification based on a hierarchical structure composed of multiple elements using a computing device, A step of generating inference result data regarding whether each element of the hierarchical structure corresponds to the given classification target data using a pre-trained inference model based on the hierarchical structure; A step of performing a Möbius transformation based on the above inference result data to calculate a plurality of transformation values corresponding to the plurality of elements; and A method comprising the step of deriving an element corresponding to the classification target data among the plurality of elements based on the plurality of transformation values.
In claim 1, In the above generating step, The method, wherein the above inference model is a pre-trained inference model using a training dataset configured to include information on one or more elements among a plurality of elements of the above hierarchical structure that correspond to a given learning target data.
In claim 1, In the above generating step, A method in which the above inference result data is composed of a vector containing multiple inference values corresponding to the multiple elements.
In claim 3, In the above-mentioned calculation step, A method for calculating a plurality of transformation values corresponding to a plurality of elements based on a plurality of inference values corresponding to a plurality of elements and a plurality of Möbius function values calculated by reflecting the hierarchical structure.
In claim 4, Multiple elements of the above hierarchical structure are, A method for constructing a partially ordered set in which any two elements have a greater or lesser relationship or an incomparable relationship.
In claim 4, In the above-mentioned calculation step, In the above hierarchical structure, when the first element (x) is the same as the second element (y), the Möbius function value becomes 1 (μ(x,y) = 1), and When the first element (x) and the second element (y) cannot be compared, the value of the Möbius function becomes 0 (μ(x,y) = 0), and The Möbius function value when the second element (y) is greater than the first element (x) is The method for this (μ(x,y) = -Σ x≤z<y μ(x,z)).
In claim 4, In the above calculation step, For a first function (f(s)) defined on a partially ordered set (S), if a second function (g(t)) is defined as the sum of the first function (g(t) = Σ s ≤ t f(s)), In the above Möbius transform, the first function (f(s)) is transformed into the sum of the second function (g(t)). ), method.
In claim 1, In the above derivation step, A method for performing hierarchical classification by deriving an element corresponding to the data to be classified based on a plurality of probability values calculated by applying softmax to the plurality of transformation values.
In claim 8, A method for performing multi-level classification by deriving two or more elements that satisfy a predetermined threshold among multiple probability values.
A method for training an inference model that performs classification based on a hierarchical structure composed of multiple elements using a computing device, A step of preparing a training dataset by labeling one or more elements corresponding to the given training target data from a plurality of elements of the above-mentioned hierarchical structure; and A method comprising the step of performing training on an inference model based on the above-mentioned training dataset.
In claim 10, In the above preparation step, A method comprising the above training dataset including the above training target data and a vector indicating whether one or more elements corresponding to the training target data among the plurality of elements are labeled.
In claim 11, Multiple elements of the above hierarchical structure are, A method for constructing a partially ordered set in which any two elements have a greater or lesser relationship or an incomparable relationship.
In claim 11, The steps performed above are, A step of generating inference result data regarding whether the training target data of the training dataset corresponds to each element of the hierarchical structure using the above inference model; A step of performing a Möbius transformation based on the above inference result data to calculate a plurality of transformation values corresponding to the plurality of elements; and A method comprising the step of performing an evaluation of the inference performance of the inference model based on the plurality of transformation values.
In claim 13, In the above generating step, A method in which the above inference result data is composed of a vector containing multiple inference values corresponding to the multiple elements.
In claim 14, In the above-mentioned calculation step, A method for calculating a plurality of transformation values corresponding to a plurality of elements based on a plurality of inference values corresponding to a plurality of elements and a plurality of Möbius function values calculated by reflecting the hierarchical structure.
In claim 15, In the output stage, In the above hierarchical structure, when the first element (x) is the same as the second element (y), the Möbius function value becomes 1 (μ(x,y) = 1), and When the first element (x) and the second element (y) cannot be compared, the value of the Möbius function becomes 0 (μ(x,y) = 0), and The Möbius function value when the second element (y) is greater than the first element (x) is The method for this (μ(x,y) = -Σ x≤z<y μ(x,z)).
In claim 15, In the above calculation step, For a first function (f(s)) defined on a partially ordered set (S), if a second function (g(t)) is defined as the sum of the first function (g(t) = Σ s ≤ t f(s)), In the above Möbius transform, the first function (f(s)) is transformed into the sum of the second function (g(t)). ), method.
A device for performing classification based on a hierarchical structure composed of a plurality of elements, comprising a processor; and memory, The above memory includes instructions configured to enable the device to implement a specific operation when executed by the above processor, and the specific operation is: Generating inference result data regarding whether each element of the hierarchical structure corresponds to the given classification target data using a pre-trained inference model based on the hierarchical structure; Performing a Möbius transformation based on the above inference result data to calculate a plurality of transformation values corresponding to the plurality of elements; and An apparatus comprising deriving an element corresponding to the classification target data among the plurality of elements based on the plurality of transformation values.
In a computer-readable storage medium storing instructions configured to enable a device that performs classification based on a hierarchical structure composed of a plurality of elements, including said processor when executed by said processor, to implement a specific operation, said specific operation is: Generating inference result data regarding whether each element of the hierarchical structure corresponds to the given classification target data using a pre-trained inference model based on the hierarchical structure; Performing a Möbius transformation based on the above inference result data to calculate a plurality of transformation values corresponding to the plurality of elements; and A computer-readable storage medium comprising deriving an element corresponding to the classification target data among the plurality of elements based on the plurality of transformation values.

Description

Method, apparatus, system and computer program for hierarchical classification using Mobius transformation The present invention relates to a hierarchical classification method, apparatus, system, and computer program utilizing a Möbius transform, and more specifically, to a hierarchical classification method, apparatus, system, and computer program utilizing a Möbius transform that can efficiently perform hierarchical classification based on a classification model learned by reflecting a hierarchical structure and a Möbius transform, and furthermore, can be extended to perform multi-layered classification. Classification techniques are methods for predicting which of the multiple types of labels a given data belongs to, and they have been widely used in various fields by being implemented based on various learning methods such as machine learning and deep learning. Here, classifying given data into labels with a hierarchical structure is called hierarchical classification, and furthermore, classifying given data into multiple labels is called multi-label classification. More specifically, in hierarchical classification, the hierarchical structure of labels can usually be given as a tree or a Directed Acyclic Graph (DAG), but when such a hierarchical structure is embedded in a learning model, the complexity increases excessively, leading to various problems such as difficulties in learning or inaccurate classification results. Furthermore, in the case of hierarchical multi-label classification, which combines hierarchical and multi-label classification, there is high demand due to its importance, but an appropriate solution to effectively implement multi-label classification has not yet been presented. Accordingly, there is a continuing demand for classification techniques that can effectively perform hierarchical classification by reflecting the hierarchical structure of complex labels and even implement multi-level classification, but an appropriate solution for this has not yet been presented. The accompanying drawings, which are included as part of the detailed description to aid in understanding the present invention, provide embodiments of the present invention and explain the technical concept of the present invention together with the detailed description. FIG. 1 is a diagram illustrating the configuration of a hierarchical classification system according to one embodiment of the present invention. FIG. 2 is a flowchart illustrating a hierarchical classification method according to one embodiment of the present invention. FIG. 3 is a diagram illustrating a hierarchical structure composed of a plurality of elements according to one embodiment of the present invention. FIG. 4 is a diagram illustrating the specific operation of a hierarchical classification system according to one embodiment of the present invention. FIG. 5 is a flowchart illustrating a learning method for an inference model for hierarchical classification according to an embodiment of the present invention. FIG. 6 is a diagram illustrating a specific flowchart of steps performed in a learning method of an inference model according to an embodiment of the present invention. FIG. 7 is a diagram illustrating a specific flowchart of a hierarchical classification method according to one embodiment of the present invention. FIG. 8 is a diagram illustrating a specific flowchart of a multi-layered classification method according to an embodiment of the present invention. FIG. 9 is a diagram illustrating the configuration of a computing device for performing a simulation of a supply chain according to an embodiment of the present invention. Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the accompanying drawings. The objects, specific advantages, and novel features of the present invention will become more apparent from the following detailed description and preferred embodiments in conjunction with the accompanying drawings. Prior to this, the terms and words used in this specification and claims are appropriately defined by the inventor to best describe his invention and should be interpreted in a meaning and concept consistent with the technical spirit of the invention; they are intended merely to describe embodiments and should not be interpreted as limiting the invention. In assigning reference numerals to components, identical or similar components are assigned the same reference numeral regardless of the reference numeral, and redundant descriptions thereof are omitted. The suffixes "module" and "part" used for components in the following description are assigned or used interchangeably for the sake of ease of drafting the specification; they do not inherently possess distinct meanings or roles and may refer to software or hardware components. In describing the components of the present invention, when a component is expressed in a singular form, it should be understood that the co