Apparatus and method for diagnosing defects using learning model based on continuous learning

Abstract

Disclosed are an apparatus and a method for diagnosing defects using a learning model based on continuous learning. A method for diagnosing a defect includes: loading, by a learning unit, a buffer learning dataset, which is learning data selected from a past learning dataset in accordance with a degree of influence on prediction performance of a current learning model during a continuous learning process; a step of learning a current learning model by using the buffer learning data set and the current learning data set; and detecting, by the detection unit, a defect from a radiation image using the current learning model when the radiation image input is received.

Description

TECHNICAL FIELD

[0001] The present application relates to a defect diagnosis technology, and more particularly, to an apparatus for diagnosing defects using a learning model based on continuous learning and a method thereof.

[0002] The present application relates to a learning data selection technology, and more particularly, to an apparatus for selecting learning data for continuous learning and a method thereof.

[0003] The present application relates to a learning data update technology, and more particularly, to an apparatus for updating learning data for continuous learning and a method thereof. BACKGROUND

[0004] Radiographic Test (RT) is a method of selecting a radiation such as X-ray and γ-ray according to a use condition and a purpose, and making the radiation penetrate a test object, thereby performing imaging on an x-ray film and detecting a defect inside the test object. It is the most widely used method among non-destructive inspection methods for detecting internal defects.

[0005] For defect diagnosis based on an artificial intelligence-based industrial radiographic test (RT), the main task is object detection, and it is necessary to maintain the accuracy of object detection as time passes. In addition, as a common method of re-learning the characteristics of a new data set generated as time passes while operating an artificial intelligence model, a transfer learning method is used, but a forgetting phenomenon in which the accuracy related to previous data decreases can occur. In the case of retaining all previous learning data sets in order to prevent the above-described problem, a storage space problem can occur. SUMMARY

[0006] The present application relates to a defect diagnosis technology, and more particularly, to an apparatus for diagnosing defects using a learning model based on continuous learning and a method thereof.

[0007] The present application relates to a learning data selection technology, and more particularly, to an apparatus for selecting learning data for continuous learning and a method thereof.

[0008] The present application relates to a learning data update technology, and more particularly, to an apparatus for updating learning data for continuous learning and a method thereof.

[0009] In order to achieve the above object, a method of diagnosing defects according to a preferred embodiment of the present application includes the steps of learning, by a learning part, a current learning model using a buffer learning data set previously stored in a buffer and a current learning data set, and detecting, by a detection part, a defect from a radiographic image using the current learning model when the radiographic image is input.

[0010] The step of learning the current learning model includes a step of loading, by the learning unit, learning data selected from past learning data sets according to degrees of influence on predictive performance of the current learning model in the continuous learning process, i.e., a buffer learning data set, and a step of learning the current learning model using the buffer learning data set and the current learning data set.

[0011] The method further includes, after the step of learning the current learning model, a step of calculating, by the switching unit, degrees of change from a plurality of past learning models of previous learning rounds that are two or more rounds apart from the current learning model to the current learning model in the continuous learning, a step of determining, by the switching unit, whether the degrees of change reach a reference value or more, a step of deciding, by the switching unit, to update the buffer learning data set in a case where the degrees of change reach the reference value or more, a step of calculating, by the updating unit, degrees of influence of a plurality of learning data in the current learning data set on predictive performance of a future learning model that is to be learned in a subsequent learning round of the current learning model when the future learning model is learned using the current learning data set in the continuous learning process, and a step of updating, by the updating unit, the buffer learning data set from a part of the plurality of learning data according to an order from high to low of the degrees of influence.

[0012] The step of calculating the degrees of influence includes a step of calculating, by the updating unit, a plasticity score indicating a probability that a predictive value of the current learning model and a predictive value of the future learning model differ for learning data in the current learning data set, a step of calculating, by the updating unit, a stability score indicating a probability that a predictive value of the current learning model and a predictive value of a first past learning model learned in a previous learning round of the current learning model, i.e., a first past learning round, differ for learning data in the current learning data set, and a step of calculating, by the updating unit, a weighted average of the plasticity score and the stability score as the degree of influence.

[0013] The step of calculating the stability score is characterized in that:

[0014] According to a mathematical formula

[0015]

[0016] The calculation is performed,

[0017] The is a stability score, the is learning data of the current learning data set, and the is a predicted value of the first past learning model, and is a predicted value of the current learning model.

[0018] The step of calculating the plasticity score, characterized in that:

[0019] According to the mathematical formula

[0020]

[0021] is calculated,

[0022] The is a plasticity score, the is learning data of a current learning dataset, the is a predicted value of the future learning model, and the is a predicted value of the current learning model.

[0023] The step of calculating the plasticity score, characterized in that:

[0024] According to the mathematical formula

[0025]

[0026] The predicted value of the future learning model is derived,

[0027] The is a predicted value of the future learning model, the is learning data of a current learning dataset, and the GP(θ t ) is a gradient vector predicted by the gradient prediction model from the future learning model.

[0028] The step of calculating the influence degree, characterized in that:

[0029] According to the mathematical formula

[0030] S i = λ · Plasticity i + (1 - λ) · Stability i

[0031] is calculated,

[0032] The S is an influence degree, the i is an index of a current learning data, the λ is a weight, the Plasticity is a plasticity score, and the Stability is a stability score.

[0033] The step of calculating the degree of similarity is characterized by deriving the degree of similarity from a first gradient vector representing a change between a weight vector of a current learning model and a weight vector of a first past learning model which has learned in a previous learning round of the first past learning model and a second gradient vector representing a change between the weight vector of the first past learning model and a weight vector of a second past learning model which has learned in a previous learning round of the first past learning model.

[0034] The degree of similarity is characterized by:

[0035] According to a mathematical expression

[0036]

[0037] The calculation is performed,

[0038] The CS is a degree of similarity, the g t is a first gradient vector between a weight vector of a current learning model and a weight vector of a first past learning model, the g t-1 is a second gradient vector between the weight vector of the first past learning model and a weight vector of a second past learning model, and the is a transposed vector of the second gradient vector.

[0039] In order to achieve the above-described object, a device for diagnosing a defect according to a preferred embodiment of the present application includes a learning part which learns a current learning model using a buffer learning data set selected from past learning data sets according to a degree of influence on a prediction performance of the current learning model in a continuous learning process and a current learning data set, and a detecting part which detects a defect from a radiological image using the current learning model when the radiological image is input.

[0040] The learning part is characterized by loading a buffer learning data set which is learning data selected from past learning data sets according to a degree of influence on a prediction performance of a current learning model in a continuous learning process and learning the current learning model using the buffer learning data set and a current learning data set.

[0041] The device further includes a switching section that calculates a degree of change from a plurality of past learning models of previous learning rounds that are two or more rounds apart from the current learning model to the current learning model in the continuous learning, determines whether the degree of change reaches a reference value or more, and decides to update the buffer learning dataset in a case where the degree of change reaches the reference value or more, and an updating section that calculates a degree of influence of a plurality of learning data in the current learning dataset on a prediction performance of a future learning model that is to be learned in a subsequent learning round of the current learning model when the future learning model is learned using the current learning dataset in the continuous learning, and selects a part of the plurality of learning data in the current learning dataset according to an order from high to low of the degrees of influence to update the buffer learning dataset.

[0042] The updating section is characterized in that the updating section calculates a plasticity score that indicates a probability that a prediction value of the current learning model and a prediction value of the future learning model differ, for the learning data in the current learning dataset, calculates a stability score that indicates a probability that a prediction value of the current learning model and a prediction value of a first past learning model that is learned in a previous learning round of the current learning model, i.e., a first past learning round, differ, for the learning data in the current learning dataset, and calculates a weighted average of the plasticity score and the stability score as the degree of influence.

[0043] The updating section is characterized in that:

[0044] According to a mathematical expression

[0045]

[0046] The stability score is calculated as

[0047] The is a stability score, the is learning data of a current learning dataset, the is a prediction value of a first past learning model, and the is a prediction value of a current learning model.

[0048] The updating section is characterized in that:

[0049] According to a mathematical expression

[0050]

[0051] The plasticity score is calculated as

[0052] The is a plasticity score, the is a prediction value of a future learning model, and the is a prediction value of a future learning model, and the is a prediction value of a current learning model.

[0053] the update unit is characterized by:

[0054] According to mathematical formula

[0055]

[0056] deriving a prediction value of the future learning model,

[0057] the update unit is characterized by: is a prediction value of a future learning model, and the is learning data of a current learning dataset, and the GP(θ t ) is a gradient vector predicted by the future learning model through the gradient prediction model.

[0058] the update unit is characterized by:

[0059] According to mathematical formula

[0060] S i = λ · Plasticity i + (1 - λ) · Stability i

[0061] calculating the degree of influence,

[0062] the S is a degree of influence, the i is an index of a current learning data, the λ is a weight, the Plasticity is a plasticity score, and the Stability is a stability score.

[0063] the switching unit is characterized by deriving the degree of change from a degree of similarity of a first gradient vector representing a change between a weight vector of a current learning model and a weight vector of a first past learning model that has learned in a previous learning round of the current learning model, and a second gradient vector representing a change between the weight vector of the first past learning model and a weight vector of a second past learning model that has learned in a previous learning round of the first past learning model.

[0064] the switching unit is characterized by:

[0065] According to mathematical formula

[0066]

[0067] calculating the similarity degree,

[0068] The CS is a similarity degree, the g t is a first gradient vector between a weight vector of a current learning model and a weight vector of a first past learning model, the g t-1 is a second gradient vector between the weight vector of the first past learning model and a weight vector of a second past learning model, and the is a transposed vector of the second gradient vector.

[0069] In order to achieve the above-mentioned object, the learning data selection method according to the preferred embodiment of the present application comprises the following steps: a step of learning, by a learning unit, a current learning model using a buffer learning data set stored in a buffer and a current learning data set; a step of calculating, by an updating unit, a degree of influence of each of a plurality of learning data in the current learning data set on a prediction performance of a future learning model to be learned in a subsequent learning round of the current learning model when the future learning model is learned using the current learning data set in a continuous learning process; and a step of updating, by the updating unit, the buffer learning data set using a part of the plurality of learning data in a descending order of the degree of influence.

[0070] The step of calculating the degree of influence comprises the following steps: a step of calculating, by the updating unit, a plasticity score representing a probability that a prediction value of the current learning model and a prediction value of the future learning model are different for each of the learning data in the current learning data set; a step of calculating, by the updating unit, a stability score representing a probability that the prediction value of the current learning model and a prediction value of a first past learning model learned in a previous learning round of the current learning model, i.e., a first past learning round, are different for each of the learning data in the current learning data set; and a step of calculating, by the updating unit, a weighted average of the plasticity score and the stability score as the degree of influence.

[0071] The step of calculating the stability score is characterized in that the updating unit calculates the stability score according to a mathematical formula

[0072]

[0073] calculating the stability score, the is a stability score, the is learning data of a current learning data set, the is a prediction value of a first past learning model, and the is a prediction value of a current learning model.

[0074] The step of calculating the plasticity score, characterized by the update unit calculating the plasticity score according to a mathematical expression

[0075]

[0076] The step of calculating the plasticity score, characterized by the update unit calculating the plasticity score according to a mathematical expression The plasticity score, characterized by the update unit calculating the plasticity score according to a mathematical expression The learning data of the current learning dataset, characterized by the learning unit constructing the learning data derived from the current learning dataset The prediction value of the future learning model, characterized by the learning unit constructing the learning data derived from the current learning dataset The prediction value of the current learning model, characterized by the learning unit constructing the learning data derived from the current learning dataset

[0077] The method, characterized by further comprising, before the step of calculating the influence degree, the step of the learning unit constructing the learning data derived from the current learning dataset, i.e., the derived learning data, and the step of the learning unit learning the gradient prediction model using the derived learning data.

[0078] The step of calculating the plasticity score, characterized by the switching unit calculating the plasticity score using the gradient prediction model.

[0079] The derived learning data, characterized by including the modified weight vector of the object layer of the current learning model each time the current learning dataset is divided in the round unit and the current learning model is iteratively learned by the learning unit, and the target gradient vector representing the difference between the weight vector of the previous round and the weight vector of the current round each time the current learning dataset is divided in the round unit and the current learning model is iteratively learned.

[0080] The step of calculating the plasticity score, characterized by the update unit,

[0081] The step of calculating the plasticity score, characterized by the update unit,

[0082]

[0083] The step of deriving the prediction value of the future learning model, characterized by the update unit, The prediction value of the future learning model, characterized by the update unit, The learning data of the current learning dataset, characterized by the update unit, t The gradient vector predicted by the gradient prediction model from the future learning model.

[0084] The step of calculating the influence degree, characterized by the update unit,

[0085] The step of calculating the influence degree, characterized by the update unit,

[0086] S i = λ · Plasticity i+ (1 - λ) • Stability i

[0087] The influence degree is calculated, S is the influence degree, i is the index of the current learning data, λ is the weight, Plasticity is the plasticity score, and Stability is the stability score.

[0088] To achieve the above-mentioned purposes, the learning data selection device according to the preferred embodiment of the present application comprises: a learning unit that learns a current learning model using a buffer learning data set stored in a buffer in advance and a current learning data set; and an updating unit that calculates degrees of influence of a plurality of learning data in the current learning data set on a prediction performance of a future learning model to be learned in a subsequent learning round of the current learning model when the future learning model is learned using the current learning data set in a continuous learning process, and updates the buffer learning data set from a part of the plurality of learning data in a descending order of the degrees of influence.

[0089] The updating unit is characterized in that the updating unit calculates, for the learning data in the current learning data set, a plasticity score representing a probability that a prediction value of the current learning model and a prediction value of the future learning model differ, calculates, for the learning data in the current learning data set, a stability score representing a probability that a prediction value of the current learning model and a prediction value of a first past learning model learned in a previous learning round of the current learning model, i.e., a first past learning round, differ, and calculates a weighted average of the plasticity score and the stability score as the influence degree.

[0090] The updating unit is characterized in that the influence degree is calculated according to a mathematical expression

[0091]

[0092] The stability score is calculated, S is the stability score, i is the index of the current learning data, λ is the weight, Plasticity is the plasticity score, and Stability is the stability score. is the stability score, i is the index of the current learning data, λ is the weight, Plasticity is the plasticity score, and Stability is the stability score. is the learning data in the current learning data set, i is the index of the current learning data, λ is the weight, Plasticity is the plasticity score, and Stability is the stability score. is the prediction value of the first past learning model, and i is the index of the current learning data. is the prediction value of the current learning model.

[0093] The updating unit is characterized in that:

[0094] The influence degree is calculated according to a mathematical expression

[0095]

[0096] The plasticity score is calculated, the plasticity score is The plasticity score is calculated, the plasticity score is The learning data of the current learning dataset is The prediction value of the future learning model is The prediction value of the current learning model is

[0097] The learning unit is characterized by constructing learning data derived from the current learning dataset, i.e., derived learning data, and learning the gradient prediction model using the constructed derived learning data, and the switching unit calculates the plasticity score using the gradient prediction model.

[0098] The derived learning data is characterized by including a modified weight vector of an object layer of the current learning model each time the learning unit divides the current learning dataset in a round unit and iteratively learns the current learning model, and a target gradient vector representing a difference between a weight vector of a previous round and a weight vector of a current round each time the current learning dataset is divided in a round unit and the current learning model is iteratively learned.

[0099] The updating unit is characterized by:

[0100] The influence degree is calculated according to a mathematical formula

[0101]

[0102] The prediction value of the future learning model is derived, the prediction value of the future learning model is The prediction value of the future learning model is derived, the prediction value of the future learning model is The learning data of the current learning dataset is t The gradient vector predicted by the gradient prediction model from the future learning model is

[0103] The updating unit is characterized by:

[0104] The influence degree is calculated according to a mathematical formula

[0105] S i = λ · Plasticity i + (1 - λ) · Stability i

[0106] The influence degree is calculated, the S is the influence degree, the i is an index of the current learning data, the λ is a weight, the Plasticity is a plasticity score, and the Stability is a stability score.

[0107] To achieve the above object, a learning data updating method according to a preferred embodiment of the present application includes: a step of learning, by a learning unit, a current learning model using a buffer learning data set stored in advance in a buffer and a current learning data set; a step of calculating, by a switching unit, a degree of change from a past learning model of a previous learning round, which is two rounds or more apart from the current learning model in a learning round of continuous learning, to the current learning model; a step of determining, by the switching unit, whether the degree of change is equal to or greater than a reference value; a step of deciding, by the switching unit, to update the buffer learning data set when the degree of change is equal to or greater than the reference value; and a step of extracting, by an updating unit, at least a part of the learning data in the current learning data set and updating the buffer learning data set.

[0108] The step of calculating the degree of change is characterized in that the switching unit derives the degree of change based on a degree of similarity between a first gradient vector representing a change between a weight vector of the current learning model and a weight vector of a first past learning model learned in a previous learning round of the current learning model and a second gradient vector representing a change between the weight vector of the first past learning model and a weight vector of a second past learning model learned in a previous learning round of the first past learning model.

[0109] The degree of similarity is characterized in that:

[0110] According to a mathematical expression

[0111]

[0112] The CS is a degree of similarity, the g t is a first gradient vector between a weight vector of the current learning model and a weight vector of the first past learning model, the g t-1 is a second gradient vector between the weight vector of the first past learning model and a weight vector of the second past learning model, and the is a transposed vector of the second gradient vector.

[0113] The step of determining whether the degree of change is equal to or greater than the reference value is characterized in that the switching unit determines that the degree of change is equal to or greater than the reference value when a degree of similarity between the first gradient vector and the second gradient vector is less than the reference value.

[0114] The method further includes, after the step of determining whether the degree of change is equal to or greater than the reference value, a step of erasing, by the switching unit, the current learning data set when the degree of change is less than the reference value.

[0115] The buffered learning data set stored in advance in the buffer is characterized by being learning data selected from past learning data sets in accordance with a degree of influence on the prediction performance of the current learning model in a continuous learning process.

[0116] To achieve the above object, a learning data updating device according to a preferred embodiment of the present application includes: a learning section that learns a current learning model using a buffered learning data set stored in advance in a buffer and a current learning data set; a switching section that calculates a degree of change from a past learning model at a previous learning round that is two or more learning rounds apart from the current learning model in a continuous learning to the current learning model, and determines to update the buffered learning data set if the degree of change is equal to or greater than a reference value after determining whether the degree of change is equal to or greater than the reference value; and an updating section that extracts at least a part of the learning data in the current learning data set and updates the buffered learning data set.

[0117] The switching section is characterized by deriving the degree of change from a degree of similarity between a first gradient vector representing a change between a weight vector of the current learning model and a weight vector of a first past learning model that learned at a previous learning round of the current learning model, and a second gradient vector representing a change between the weight vector of the first past learning model and a weight vector of a second past learning model that learned at a previous learning round of the first past learning model.

[0118] The switching section is characterized by:

[0119] According to mathematical expression

[0120]

[0121] The degree of similarity is calculated, the CS is a degree of similarity, the g t is a first gradient vector between a weight vector of the current learning model and a weight vector of the first past learning model, the g t-1 is a second gradient vector between the weight vector of the first past learning model and a weight vector of the second past learning model, and the is a transposed vector of the second gradient vector.

[0122] The switching section is characterized by determining that the degree of change is equal to or greater than the reference value if the degree of similarity between the first gradient vector and the second gradient vector is less than the reference value.

[0123] The switching section is characterized by erasing the current learning data set by the switching section if the degree of change is less than the reference value.

[0124] the buffer learning dataset stored in advance into the buffer, is characterized by being learning data selected from past learning datasets according to a degree of influence on the prediction performance of the future learning model in the continuous learning process.

[0125] The present application can minimize storage space while overcoming the catastrophic forgetting phenomenon, and thus continuously provide a learning model with high accuracy at the lowest cost, by updating the buffer learning dataset according to the prediction performance of the future learning model and simultaneously performing continuous learning of the learning model using the buffer learning dataset and the new learning dataset. BRIEF DESCRIPTION OF DRAWINGS

[0126] Figure 1 is a schematic diagram for explaining the configuration of a system for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present application.

[0127] Figure 2 is a schematic diagram for explaining the configuration of an apparatus for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present application.

[0128] Figure 3 is a schematic diagram for explaining learning datasets and learning models according to learning rounds in a continuous learning process according to an embodiment of the present application.

[0129] Figure 4 is a flowchart for explaining a selection method of learning data for continuous learning according to an embodiment of the present application.

[0130] Figure 5 is a flowchart for explaining a learning method of a gradient prediction model according to an embodiment of the present application.

[0131] Figure 6 is a flowchart for explaining a calculation method of a degree of influence of learning data on the prediction performance of a future learning model according to an embodiment of the present application.

[0132] Figure 7 is a flowchart for explaining a method of diagnosing defects using a learning model based on continuous learning according to an embodiment of the present application.

[0133] Figure 8 is a screen example for explaining a method of diagnosing defects using a learning model based on continuous learning according to an embodiment of the present application.

[0134] Figure 9 is a schematic diagram illustrating a computing device according to an embodiment of the present application. DETAILED DESCRIPTION

[0135] The present application can be modified in various ways and have various embodiments, and specific embodiments will be illustrated and described in detail below. However, this is not intended to limit the present application to specific embodiments, but should be understood to include all modifications, equivalents, and alternatives included in the spirit and scope of the present application.

[0136] The terms used in the present application are used only to describe specific embodiments, and are not intended to limit the present application. Unless the context clearly indicates otherwise, singular statements also include plural meanings. In the present application, terms such as "include" or "have" are used only to indicate that the features, numbers, steps, actions, components, parts, or combinations described in the specification are present, and should not be understood as excluding the possibility of one or more other features, numbers, steps, actions, components, parts, or combinations being present or being added.

[0137] In particular, the terms or words used in the specification and claims described below should not be interpreted as limited to the general or dictionary meanings, but should be interpreted based on the concept of the terms appropriately defined by the inventors to conform to the technical idea of the present application, and should be interpreted in the meaning and concept conforming to the technical idea of the present application. In particular, in the embodiments of the present application, "estimation" or "inference" means deriving a result calculated by a learning model (LM: Machine Learning Model / Deep Learning Model) according to learning content.

[0138] First, a system for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present application will be described. Figure 1 is a schematic diagram for explaining the configuration of a system for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present application. Figure 2 is a schematic diagram for explaining the configuration of an apparatus for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present application. Figure 3 is a schematic diagram for explaining learning data sets and learning models according to learning rounds in a continuous learning process according to an embodiment of the present application.

[0139] Referring to Figure 1 , a system for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present application includes a radiography apparatus RTA, a scanner SC, an inspection apparatus 10, and a storage apparatus 40.

[0140] The radiography apparatus RTA can perform radiography on an object body to which an image quality indicator (IQI) is attached, thereby obtaining a radiographic film on which the object body to which the image quality indicator is attached is photographed. The object body can be exemplified by a pipe, a hose, and the like.

[0141] The scanner SC can scan the radiographic film, thereby generating a digitized radiographic image. After the digitized radiographic image is generated in the manner described above, the generated radiographic image is input to the inspection apparatus 10. At this time, the radiographic image can be directly input from the scanner SC, or can be input from a corresponding storage medium after being stored in the storage medium.

[0142] The inspection apparatus 10 can learn a current learning model using a buffer learning dataset stored in the buffer area BF in advance and a current learning dataset by continuous learning, and detect a defect from the radiographic image using the current learning model that is the latest learning model among the learning models learned by the continuous learning. In addition, the inspection apparatus 10 can output a report including the radiographic image in which the defect is detected by a bounding box (BB).

[0143] The storage apparatus 40 can be a cloud server or a database server. The buffer area BF of the storage apparatus 40 can store data, and the buffer learning dataset according to an embodiment of the present application can be stored in the buffer area BF as described above. The buffer learning dataset is learning data selected from past learning datasets used in a learning process of a past learning round based on a current learning round, according to a degree of influence on a prediction performance of a current learning model in a continuous learning process. In addition, although a case in which the buffer area BF is a storage medium included in the storage apparatus 40 is described, in other embodiments of the present application, the buffer area BF can be a storage medium included in the inspection apparatus 10.

[0144] Referring to Figure 2 , the inspection apparatus 10 includes a collection part 100, a learning part 200, a switching part 300, an update part 400, and a detection part 500.

[0145] The collection part 100 can continuously collect learning data and construct a current learning dataset for continuous learning. The collection part 100 provides the constructed current learning dataset to the learning part 200 after constructing the current learning dataset.

[0146] The learning unit 200 can generate a learning model through learning (Deep Learning or Machine Learning). The learning model according to the embodiment of the present application can detect a learned object from an image. The learning model according to the embodiment of the present application detects a learned object, i.e., a defect, from a radiographic image. At this time, the learning model can detect an area occupied by the defect from the radiographic image through a bounding box (BB). The learning model as described above can be exemplified by a convolutional neural network (CNN), YOLO, a region convolutional neural network (RCNN), and a faster region convolutional neural network (Faster RCNN), etc. The learning model includes a plurality of layers (or modules) connected to each other, and the plurality of layers (or modules) are composed of a plurality of operations. In addition, the plurality of layers (or modules) are connected by weights (W). That is, an operation result output from a certain layer (or module) is input to an operation of a next layer after applying a weight. The learning model performs a plurality of operations applying weights between a plurality of layers (or modules) on input data and generates an output. In other words, a plurality of operations connected by weights between a plurality of layers (or modules) are performed. Next, the plurality of operations connected by weights between a plurality of layers (or modules) of the learning model as described above are referred to as "weight operations".

[0147] In particular, the learning unit 200 according to the embodiment of the present application learns a learning model through continuous learning.

[0148] Referring to Figure 3 , the continuous learning in the embodiment of the present application refers to a method of continuously collecting new learning data and learning a learning model using learning data sets of different learning rounds after constructing learning data sets including a plurality of learning data by dividing the continuously collected learning data by learning rounds. In the embodiment of the present application, the learning rounds can be divided according to at least one of time, the number of collected learning data, and a specific event. Thereby, the present application can continuously learn according to different learning rounds using learning data sets including a plurality of learning data divided by learning rounds, and thus can continuously learn a learning model, thereby continuously generating a learning model different from a learning model derived in a previous learning round.

[0149] In an embodiment of the present invention, when the reference time point is referred to as the current learning round (t), the learning model to be learned in the current learning round (t) is referred to as the current learning model. Furthermore, the learning dataset used to learn the current learning model is referred to as the current learning dataset. The current learning dataset includes multiple learning data collected from the previous learning round (t-1), i.e., the first past learning round (t-1), up to the current learning round (t).

[0150] Furthermore, the learning model that will be learned in the previous learning rounds of the current learning model, i.e., the first past learning round (t-1), is called the first past learning model. Additionally, the learning dataset used to learn from the first past learning model is called the first past learning dataset. The first past learning dataset includes multiple learning data collected from the previous learning rounds of the first past learning round (t-1), i.e., the second past learning round (t-2), up to the first past learning round (t-1).

[0151] Furthermore, the learning model of all previous learning epochs (t-1, t-2, ...) based on the current learning epoch (t) is referred to as the past learning model. Additionally, the learning dataset of all past learning epochs (t-1, t-2, ...) based on the current learning epoch (t) is referred to as the past learning dataset.

[0152] Furthermore, the learning model that will be learned in the next learning epoch (t+1) after the current learning model is called the future learning model. The learning dataset used to learn the future learning model is called the future learning dataset. The future learning dataset includes multiple learning data collected from the current learning epoch (t) to the future learning epoch (t+1).

[0153] Revisit Figure 2 The switching unit 300 is used to determine whether to update the buffered learning dataset according to an embodiment of the present invention and the timing of the update.

[0154] The switching unit 300 derives the degree of change from multiple past learning models that are more than two learning epochs away from the current learning model in a continuous learning cycle up to the current learning model. To this end, the switching unit 300 derives a first gradient vector representing the change between the weight vector of the current learning model and the weight vector of a first past learning model learned in a previous learning epoch of the current learning model. Next, the switching unit 300 generates a second gradient vector representing the change between the weight vector of the first past learning model and the weight vector of a second past learning model learned in a previous learning epoch of the first past learning model. Next, the switching unit 300 derives the degree of change based on the similarity between the first gradient vector and the second gradient vector. Next, the switching unit 300 determines whether the degree of change reaches or exceeds a benchmark value. That is, the switching unit 300 can determine that the degree of change reaches or exceeds a benchmark value if the similarity between the first gradient vector and the second gradient vector is less than a benchmark value, and determine that the degree of change is less than a benchmark value if the similarity reaches or exceeds a benchmark value. Based on the determination results described above, if the degree of change reaches or exceeds the baseline value (but the degree of similarity is less than the baseline value), the switching unit 300 decides to update the buffered learning dataset. Conversely, if the degree of change is less than the baseline value (but the degree of similarity reaches or exceeds the baseline value), the switching unit 300 erases the current learning dataset.

[0155] The update unit 400 calculates the degree to which each of the multiple learning data points in the current learning dataset affects the predictive performance of the future learning model. Specifically, the update unit 400 calculates the degree to which each of the multiple learning data points in the current learning dataset affects the predictive performance of the future learning model when the current learning dataset is used to learn the model for subsequent learning rounds of the current learning model (i.e., the future learning model) during continuous learning. Furthermore, the update unit 400 selects a portion of the multiple learning data points in the current learning dataset, in descending order of the calculated degree of influence, to update the buffered learning dataset. At this time, the update unit 400 deletes the buffered learning data previously stored in the buffer BF and stores the learning data selected from the multiple learning data points in descending order of influence into the buffer, thereby updating the buffered learning dataset. Additionally, the update unit 400 erases the remaining learning data from the multiple learning data points that were not selected in descending order of influence.

[0156] Next, a method for selecting learning data for continuous learning according to an embodiment of the present invention will be described. Figure 4 A method for selecting learning data for continuous learning according to an embodiment of the present invention has been described.

[0157] See Figure 4In step S110, the collection unit 100 continuously collects learning data and constructs the current learning dataset.

[0158] In step S120, after loading the buffered learning dataset that has been pre-stored in the buffer BF, the learning unit 200 generates the current learning model by learning using the buffered learning dataset pre-stored in the buffer BF and the current learning dataset. The buffered learning dataset pre-stored in the buffer BF is learning data selected from past learning datasets used in previous learning rounds based on the degree to which it affects the predictive performance of the current learning model during continuous learning.

[0159] In step S130, the switching unit 300 calculates the degree of change from multiple past learning models that are more than two learning rounds away from the current learning model in the continuous learning rounds up to the current learning model. Next, step S130 as described above will be explained in more detail.

[0160] First, the switching unit 300 derives a first gradient vector representing the change between the weight vector of the current learning model and the weight vector of the first past learning model that was learned in the previous learning rounds of the current learning model.

[0161] Next, the switching unit 300 derives a second gradient vector representing the change between the weight vector of the first past learning model and the weight vector of the second past learning model learned in previous learning rounds of the first past learning model. Next, the switching unit 300 derives the degree of change based on the similarity between the first gradient vector and the second gradient vector.

[0162] At this point, the switching unit 300 can calculate the similarity according to the following mathematical formula 1.

[0163]

Mathematical Formula 1

[0164]

[0165] Here, CS represents the degree of similarity. t This is the first gradient vector between the weight vector of the current learning model and the weight vector of the first past learning model. Furthermore, g... t-1 This is the second gradient vector between the weight vector of the first past learning model and the weight vector of the second past learning model. Furthermore, It is the transpose of the second gradient vector.

[0166] Next, in step S140, the switching unit 300 determines whether the degree of change reaches or exceeds a reference value. At this time, according to Mathematical Formula 1, if the similarity between the first gradient vector and the second gradient vector is less than the reference value, the switching unit 300 determines that the degree of change reaches or exceeds the reference value. Conversely, according to Mathematical Formula 1, if the similarity between the first gradient vector and the second gradient vector reaches or exceeds the reference value, the switching unit 300 determines that the degree of change is less than the reference value.

[0167] Based on the determination result of step S140, if the degree of change is insufficient (the degree of similarity reaches or exceeds the benchmark value), the switching unit 300 erases the current learning dataset by executing step S150, and repeats steps S110 to S140 as described above.

[0168] Conversely, based on the determination result of step S140, if the degree of change reaches or exceeds the baseline value (but the degree of similarity is less than the baseline value), it is decided to update the buffer learning dataset and execute step S160.

[0169] In step S160, the learning unit 200 learns the gradient prediction model (GPM). The gradient prediction model (GPM) can collect derived learning data from the learning data used to learn the current learning model, and use the collected derived learning data to predict the gradient vector of the future learning model. The learning of the gradient prediction model (GPM) will be explained in more detail later.

[0170] Next, in step S170, the update unit 400 calculates the degree to which each of the multiple learning data points in the current learning dataset affects the predictive performance of the future learning model. Specifically, it calculates the degree to which each of the multiple learning data points in the current learning dataset affects the predictive performance of the future learning model when the current learning dataset is used to learn the model for subsequent learning rounds of the current learning model (i.e., the future learning model) during continuous learning. The step of calculating the degree of influence as described above will be explained in more detail later.

[0171] Next, in step S180, the update unit 400 selects a portion of the multiple learning data from the current learning dataset in descending order of influence to update the buffered learning dataset. At this time, the update unit 400 deletes the buffered learning data previously stored in the buffer BF and stores the learning data selected from the multiple learning data in descending order of influence into the buffer BF, thereby updating the buffered learning dataset. Furthermore, in step S180, the remaining unselected learning data from the multiple learning data in descending order of influence is erased.

[0172] By updating the buffered learning dataset as described above and simultaneously performing continuous learning of the learning model using both the buffered learning dataset and the new learning dataset, storage space can be minimized while overcoming catastrophic forgetting, thereby continuously providing a learning model that achieves high accuracy at the lowest cost.

[0173] Next, the learning method of the gradient prediction model (GPM) according to an embodiment of the present invention will be described. Figure 5 This is a flowchart illustrating a learning method for a gradient prediction model according to an embodiment of the present invention. Specifically, Figure 5 Yes Figure 4 Detailed explanation of step S160.

[0174] See Figure 5 In step S210, the learning unit 200 is in a state of selecting the object layer of the current learning model. For example, it is assumed that the object layer is a convolutional layer. Furthermore, in step S210, while learning the current learning model in step 120, the learning unit 200 divides and learns multiple learning data in the current learning dataset in epoch units, and stores the modified weight vector of the object layer of the current learning model each time it is divided and learned in epoch units.

[0175] In this way, the learning unit 200 constructs multiple derived learning data in step S220, which are derived from the current learning dataset.

[0176] The derived learning data includes the modified weight vector of the object layer of the current learning model when the current learning dataset is divided into epoch units and the current learning model is iterated, and the target gradient vector representing the difference between the weight vector of the previous epoch and the weight vector of the current epoch when the current learning dataset is divided into epoch units and the current learning model is iterated.

[0177] Each time multiple learning data are divided into epoch units and iterated, the modified weight vector of the object layer will change in the manner shown in Equation 2 below.

[0178]

Mathematical Formula 2

[0179] W(t+1)=W(t)+LR×G(t)

[0180] Where t is the index of the round. LR represents the learning rate. In addition, G(t) represents the gradient vector. This invention approximates the product of the learning rate LR and the gradient vector G(t) into a gradient vector. Thus, the weight vector can be represented as [W(1), W(2), W(3), ..., W(E-1)] (E is an arbitrarily positive constant). In addition, the target gradient vector corresponding to [W(1), W(2), W(3), ..., W(E-1)] can be represented as [W(2)-W(1), [W(3)-W(2), [W(4)-W(3), ..., [W(E)-W(E-1)]].

[0181] Next, in step S230, the learning unit 200 inputs the weight vector from the derived learning data into the gradient prediction model (GPM) which has weights that have not yet been learned. Then, in step S240, the gradient prediction model (GPM) performs a weight calculation applying the unlearned weights to the radiographic images used for learning, thereby generating a weight difference vector that predicts the difference between the weight vectors from previous rounds and the weight vector of the object layer in the current round.

[0182] Next, in step S250, the learning unit 200 calculates the loss representing the difference between the target gradient vector and the weight difference vector using a loss function. Then, in step S260, the learning unit 200 performs optimization by modifying the weights of the gradient prediction model (GPM) to minimize the loss derived from the loss function.

[0183] Next, the learning unit 200 determines in step S270 whether the learning completion condition is met. In one embodiment, the learning completion condition may be that the loss calculated in step S250 converges and reaches a predetermined target value or lower. Based on the determination result of step S270 as described above, if the learning completion condition is not met, steps S230 to S270 as described above are repeated using different learning data. Conversely, based on the determination result of step S270, if the learning completion condition is met, the learning unit 200 completes the learning of the gradient prediction model (GPM) in step S280.

[0184] Next, a method for calculating the degree to which multiple learning data in the current learning dataset, according to an embodiment of the present invention, affect the predictive performance of the future learning model will be described. Figure 6 This is a flowchart illustrating a method for calculating the degree to which learning data, according to embodiments of the present invention, affects the predictive performance of a future learning model. That is, Figure 6 Yes Figure 4 Detailed explanation of step S170.

[0185] In step S310, the update unit 400 calculates a plasticity score for the learning data in the current learning dataset, which represents the probability that the predicted value of the current learning model will differ from the predicted value of the future learning model.

[0186] At this point, the update unit 400 can calculate the plasticity fraction according to the following mathematical formula 3.

[0187]

Mathematical Expression 3

[0188]

[0189] in, Let represent the conditional probability of the plasticity score. n is the index of the learning data in the current learning dataset. Superscripts such as t represent the index of the learning epoch in the continuous learning process (and can also be used as the index of the learning model), where t represents the current learning epoch and t-1 represents the first past learning epoch. This refers to the learning data in the current learning dataset. Furthermore, This is the predicted value of the future learning model, and This represents the predicted value of the current learning model.

[0190] Furthermore, since the predictions of the future learning model are made before the future learning model was generated, the prediction gradient of the future learning model is estimated using the gradient prediction model (GPM). Therefore, the predictions of the future learning model in Equation 3 will be estimated according to Equation 4 below.

[0191]

Mathematical Expression 4

[0192]

[0193] in, This refers to the learning data in the current learning dataset. Furthermore, The predicted value of the future learning model, and the GP(θ) t ) represents the gradient vector used by the Gradient Prediction Model (GPM) to predict the future learning model.

[0194] Next, in step S320, the update unit 400 calculates a stability score for the learning data in the current learning dataset, which represents the probability that the predicted value of the current learning model differs from the predicted value of the first past learning model that was learned in the previous learning rounds of the current learning model, i.e., the first past learning rounds.

[0195] At this point, the update unit 400 can calculate the stability score according to the following mathematical formula 5.

[0196]

Mathematical Expression 5

[0197]

[0198] in, Let represent the conditional probability of the stability score.

[0199] n is the index of the learning data in the current learning dataset. Furthermore, superscripts such as t are indices of the learning epochs in the continuous learning process, where t represents the current learning epoch and t-1 represents the first past learning epoch. This represents the learning data in the current learning dataset. Furthermore, The predicted value is the first past learning model, and the... This represents the predicted value of the current learning model.

[0200] Next, in step S330, the update unit 400 calculates the weighted average of the plasticity score and the stability score as the degree of influence.

[0201] At this point, the update unit 400 can calculate the degree of influence according to the following mathematical formula 6.

[0202]

Mathematical Expression 6

[0203] S i =λ·Plasticity i +(1-λ)·Stability i

[0204] Where S represents the degree of influence. i is the index of the learning data in the current learning dataset. λ is the pre-defined weight. Plasticity represents the plasticity score, while Stability represents the stability score.

[0205] Next, a method for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present invention will be described. Figure 7 This is a flowchart illustrating a method for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present invention. Figure 8 This is a screen example used to illustrate a method for diagnosing defects using a learning model based on continuous learning according to an embodiment of the present invention.

[0206] exist Figure 7 In the above, the learning model, as described in step S120, is the current learning model that learns using a pre-stored buffered learning dataset and the current learning dataset.

[0207] See Figure 7In step S410, the radiographic imaging device (RTA) performs radiographic imaging on an object equipped with an Image Quality Indicator (IQI), thereby generating a radiographic penetration inspection film of the object. The object can be, for example, a pipe or hose.

[0208] Next, the scanner SC scans the radiographic examination film in step S420 to generate a digitized radiographic image.

[0209] After generating the digitized radiographic image as described above, the generated radiographic image will be input into the inspection apparatus 10. At this time, the radiographic image can be directly input from the scanner SC, or it can be input from the corresponding storage medium after being stored in another storage medium.

[0210] In step S430, the detection unit 500 of the inspection device 10 receives the input of a radiographic image. Next, in step S440, the detection unit 500 detects defects from the radiographic image using the latest model from the continuously learned learning model, i.e., the current learning model, as described above. At this time, the current learning model is as follows: Figure 8 As shown, the area occupied by a defect can be detected from a radiographic image using bounding boxes (BBs).

[0211] Next, in step S450, the inspection unit 500 outputs a report including radiographic images of defects detected by bounding boxes (BBs).

[0212] Figure 9 This is a schematic diagram illustrating a computing device according to an embodiment of the present invention. Figure 9 The computing device TN100 in this specification may be any device described herein (e.g., the checking device 10 and the storage device 40, etc.).

[0213] exist Figure 9 In some embodiments, the computing device TN100 may include at least one processor TN110, a transceiver device TN120, and a memory TN130. Furthermore, the computing device TN100 may also include a storage device TN140, an input interface device TN150, and an output interface device TN160, etc. The components included in the computing device TN100 can be connected and communicate with each other via a bus TN170.

[0214] The processor TN110 can execute program commands stored in at least one of the memory TN130 and the storage device TN140. The processor TN110 can refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor executing the methods according to embodiments of the present invention. The processor TN110 can be configured to implement the steps, functions, and methods described in conjunction with embodiments of the present invention. The processor TN110 can control the various components of the computing device TN100.

[0215] The memory TN130 and the storage device TN140 can store various types of information related to the operation of the processor TN110. The memory TN130 and the storage device TN140 can be composed of at least one of volatile storage media and non-volatile storage media. For example, the memory TN130 can be composed of at least one of read-only memory (ROM) and random access memory (RAM).

[0216] The transceiver TN120 can transmit or receive wired or wireless signals. The transceiver TN120 can also perform communication by connecting to a network.

[0217] In particular, the collection unit 100, learning unit 200, switching unit 300, updating unit 400, and detection unit 500 of the inspection device 10 according to an embodiment of the present invention can be implemented in a computer-readable program form and stored in the memory TN130, and then run by the processor TN110. Alternatively, the collection unit 100, learning unit 200, switching unit 300, updating unit 400, and detection unit 500 can be subordinate modules of the processor TN110.

[0218] Furthermore, according to an embodiment of the present invention, the buffer BF may be the memory TN130 or the storage device TN140 of the inspection device 10 or the storage device 40.

[0219] Furthermore, the various methods described above according to embodiments of the present invention can be implemented in a program form readable by various computer components and recorded on a computer-readable recording medium. The recording medium may individually contain program instructions, data files, and data structures, or a combination thereof. The program instructions recorded on the recording medium may be specially designed and configured for the present invention, or may be known and usable by those skilled in the art of computer software. For example, the recording medium includes magnetic media such as hard disks, floppy disks, and magnetic tapes; optical media such as CD-ROMs and DVDs; magneto-optical media such as floppy disks; read-only memory (ROM); random access memory (RAM); and flash memory, etc., hardware devices specifically configured to store and execute program instructions. Examples of program instructions may include not only machine code generated by a compiler, but also high-level language code that runs on a computer using tools such as an interpreter. The hardware device described above can be configured to operate as more than one software module to perform the actions of the present invention, and vice versa.

[0220] In the foregoing, one embodiment of the present invention has been described. However, those skilled in the art can make various modifications and alterations to the present invention without departing from the spirit of the invention as set forth in the claims by adding, changing, deleting, or supplementing the constituent elements. Such modifications and alterations should be understood to be included within the scope of the claims of the present invention.

Claims

1. A method for diagnosing defects, characterized in that, include: The steps involve the learning department using a buffered learning dataset stored in a buffer beforehand and the current learning dataset to learn the current learning model; as well as, The step of the detection unit detecting defects from radiographic images using the current learning model upon receiving the input of radiographic images.

2. The method for diagnosing defects according to claim 1, characterized in that: The steps for learning the current learning model include: The step of loading the buffered learning dataset, which is selected from past learning datasets during continuous learning based on its impact on the predictive performance of the current learning model, by the learning unit; and, The steps of learning the current learning model using the buffered learning dataset and the current learning dataset.

3. The method for diagnosing defects according to claim 1, characterized in that: Following the step of learning the current learning model, the method further includes: The step of calculating the degree of change of multiple past learning models from previous learning rounds that are more than two rounds away from the current learning model in the continuous learning rounds to the current learning model; The step of determining whether the degree of change reaches or exceeds a reference value by the switching unit; The step of the switching unit deciding to update the buffered learning dataset when the degree of change reaches or exceeds the benchmark value; The step of calculating, by the update unit, the degree to which multiple learning data points in the current learning dataset affect the predictive performance of the future learning model when learning a future learning model to be learned in subsequent learning rounds of the current learning model using the current learning dataset during continuous learning; and, The step of updating the buffered learning dataset by selecting a portion of the plurality of learning data in descending order of the degree of influence.

4. The method for diagnosing defects according to claim 3, characterized in that: The steps for calculating the extent of the influence include: The step of the updating unit calculating a plasticity score, which represents the probability that the predicted value of the current learning model differs from the predicted value of the future learning model, based on the learning data in the current learning dataset; The steps include: calculating a stability score, representing the probability that the predicted value of the current learning model differs from the predicted value of a first past learning model learned in previous learning rounds, using the learning data in the current learning dataset; and... The step of calculating the weighted average of the plasticity score and the stability score by the updating unit as the degree of influence.

5. The method for diagnosing defects according to claim 4, characterized in that: In the step of calculating the stability score, According to the mathematical formula Perform calculations. The The stability score is... The The learning data for the current learning dataset. The The predicted value is the value from the first past learning model. The This is the predicted value of the current learning model.

6. The method for diagnosing defects according to claim 4, characterized in that: In the step of calculating the plasticity fraction According to the mathematical formula Perform calculations. The The plasticity fraction, The The learning data for the current learning dataset. The The predicted value of the future learning model. The This is the predicted value of the current learning model.

7. The method for diagnosing defects according to claim 6, characterized in that: In the step of calculating the plasticity fraction According to the mathematical formula The predicted value of the future learning model is derived. The The predicted value of the future learning model. The The learning data for the current learning dataset. The GP(θ) t ) is the gradient vector predicted by the future learning model through the gradient prediction model.

8. The method for diagnosing defects according to claim 3, characterized in that: In the step of calculating the extent of the influence, According to the mathematical formula S i =λ·Plasticity i +(1-λ)·Stability i Perform calculations. S represents the degree of the influence. i is the index of the current learning data in the current learning dataset. λ is the weight. The Plasticity score is the plasticity score. The Stability refers to the stability score.

9. The method for diagnosing defects according to claim 3, characterized in that: In the step of calculating the degree of change The degree of change is derived by comparing the similarity between a first gradient vector representing the change between the weight vector of the current learning model and the weight vector of a first past learning model learned in a previous learning round of the current learning model, and a second gradient vector representing the change between the weight vector of the first past learning model and the weight vector of a second past learning model learned in a previous learning round of the first past learning model.

10. The method for diagnosing defects according to claim 9, characterized in that: The degree of similarity is based on the mathematical formula Perform calculations. The CS refers to the degree of similarity. The g t The first gradient vector is the gradient between the weight vector of the current learning model and the weight vector of the first past learning model. The g t-1 The second gradient vector is the gradient between the weight vector of the first past learning model and the weight vector of the second past learning model. The is the transpose of the second gradient vector.

11. A device for diagnosing defects, characterized in that, include: The learning department uses a buffered learning dataset stored in a buffer beforehand, along with the current learning dataset, to learn the current learning model; as well as, The detection unit, upon receiving input of a radiographic image, uses the current learning model to detect defects from the radiographic image.

12. The apparatus for diagnosing defects according to claim 11, characterized in that: The learning department The buffered learning dataset is the learning data selected from past learning datasets during continuous learning based on the degree to which it affects the predictive performance of the current learning model. The current learning model is learned using the buffered learning dataset and the current learning dataset.

13. The apparatus for diagnosing defects according to claim 11, characterized in that, Also includes: The switching unit calculates the degree of change from multiple past learning models that are more than two learning epochs away from the current learning model in the continuous learning epochs up to the current learning model. Determine whether the degree of change reaches or exceeds the benchmark value. If the degree of change reaches or exceeds the baseline value, it is decided to update the buffered learning dataset; and, The update unit calculates when, during continuous learning, it uses the current learning dataset to learn a future learning model that will be learned in subsequent learning rounds of the current learning model. The degree to which each of the multiple learning data points in the current learning dataset affects the predictive performance of the future learning model. The buffered learning dataset is updated by selecting a portion of the plurality of learning data in descending order of the degree of influence.

14. The apparatus for diagnosing defects according to claim 13, characterized in that: The update section For the learning data in the current learning dataset, a plasticity score is calculated, representing the probability that the predicted value of the current learning model differs from the predicted value of the future learning model. For the learning data in the current learning dataset, a stability score is calculated, representing the probability that the predicted value of the current learning model differs from the predicted value of a first past learning model that was learned in the previous learning rounds of the current learning model. The weighted average of the plasticity score and the stability score is calculated as the degree of influence.

15. The apparatus for diagnosing defects according to claim 14, characterized in that: The updating unit is based on the mathematical formula Calculate the stability score. The The stability score is... The The learning data for the current learning dataset. The The predicted value is the value from the first past learning model. The This is the predicted value of the current learning model.

16. The apparatus for diagnosing defects according to claim 14, characterized in that: The update section According to the mathematical formula Calculate the plasticity fraction. The The plasticity fraction, The The learning data for the current learning dataset. The The predicted value of the future learning model. The This is the predicted value of the current learning model.

17. The apparatus for diagnosing defects according to claim 16, characterized in that: The updating unit is based on the mathematical formula The predicted value of the future learning model is derived. The The predicted value of the future learning model. The The learning data for the current learning dataset. The GP(θ) t ) is the gradient vector predicted by the future learning model through the gradient prediction model.

18. The apparatus for diagnosing defects according to claim 13, characterized in that: The updating unit is based on the mathematical formula S i =λ·Plasticity i +(1-λ)·Stability i Calculate the extent of the impact. S represents the degree of the influence. i is the index of the current learning data in the current learning dataset. λ is the weight. The Plasticity score is the plasticity score. The Stability refers to the stability score.

19. The apparatus for diagnosing defects according to claim 13, characterized in that: The switching unit The degree of change is derived by comparing the similarity between a first gradient vector representing the change between the weight vector of the current learning model and the weight vector of a first past learning model learned in a previous learning round of the current learning model, and a second gradient vector representing the change between the weight vector of the first past learning model and the weight vector of a second past learning model learned in a previous learning round of the first past learning model.

20. The apparatus for diagnosing defects according to claim 19, characterized in that: The switching unit is based on the mathematical formula Calculate the degree of similarity. The CS refers to the degree of similarity. The g t The first gradient vector is the gradient between the weight vector of the current learning model and the weight vector of the first past learning model. The g t-1 The second gradient vector is the gradient between the weight vector of the first past learning model and the weight vector of the second past learning model. The is the transpose of the second gradient vector.

Images