A method for analyzing data based on level of data lens and a computing device on which the method is implemented

The computing device addresses inefficiencies in data quality analysis by enabling precise visualization and interaction, enhancing data exploration and quality improvement through user-driven data processing and visualization.

KR103043798B1Active Publication Date: 2026-11-10PEBLUS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
KR1020250019944
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Priority Date
2024-08-16
Filing Date
2025-02-17
Publication Date
2026-11-10
Estimated Expiration
2045-02-17

AI Technical Summary

Technical Problem

Existing data quality diagnostic methods struggle with accurately analyzing unstructured data and visualizing high-dimensional datasets, leading to inefficient data exploration and quality improvement processes due to the lack of intuitive interaction and effective reflection of data structures.

Method used

A computing device equipped with a processor and memory, configured to execute instructions for data processing, selects a data processing model based on user input, provides prior information through a GUI, constructs a data processing model, and visualizes the data in two or three dimensions using vector sets, allowing for intuitive interaction and visualization.

Benefits of technology

Enables automatic diagnosis and intuitive visualization of large datasets, facilitating better understanding of data structure and improving data quality for machine learning model performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112025017825037-PAT00008_ABST
    Figure 112025017825037-PAT00008_ABST
Patent Text Reader

Abstract

According to one embodiment of the present disclosure, a computing device may be provided that includes a memory and at least one processor electronically connected to the memory, and is configured to execute at least one instruction stored in the memory, wherein the processor is configured to perform the operation of acquiring a first data set, the operation of selecting a first level among a plurality of levels classified according to a data processing method based on user input, the operation of determining at least one attribute of a first data processing model corresponding to the first level—the first data processing model includes at least one artificial intelligence model—the operation of providing prior information related to the task of processing the first data set using the first data processing model through a first GUI (Graphic User Interface), and the operation of constructing the first data processing model based on user input to the first GUI.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present disclosure relates to a data processing technology for diagnosing and visualizing data. More specifically, it relates to a technology that diagnoses data through imaging and provides interaction functions by visualizing the data. Background Technology

[0003] With the recent widespread adoption of Artificial Intelligence (AI) and Machine Learning (ML) technologies across various industries, the importance of techniques for effectively processing and analyzing large-scale datasets is increasing. Data quality is a critical factor that directly impacts the performance of AI models (or neural networks), and vectorization and visualization techniques are essential for diagnosing and analyzing this data. In particular, as large volumes of images, text, and structured data are used to train deep learning-based models, it is crucial to accurately understand the inherent characteristics of the data.

[0004] Existing data quality diagnostic methods rely primarily on integrity verification and basic statistical analysis of structured data, making it difficult to perform precise analysis of unstructured data. Furthermore, existing technologies for identifying the implicit distribution of large-scale datasets fail to effectively reflect data structures in high-dimensional spaces and lack the functionality to provide a visual understanding when evaluating data quality.

[0005] Existing data visualization techniques also have limitations. In the case of high-dimensional data, it is difficult to maintain the characteristics of the original data even when dimensionality reduction techniques are used to represent it in a 2D or 3D space. Furthermore, the lack of intuitive interaction during the process of users analyzing and manipulating the quality of data sets leads to inefficient data exploration and quality improvement processes. The problem to be solved

[0007] One objective of the present disclosure is to diagnose the quality of a large dataset.

[0008] In addition, one objective of the present disclosure is to visualize a dataset so that it can be effectively represented while maintaining the intrinsic structure of the data through a vectorization method for precisely identifying the intrinsic distribution of a large dataset.

[0009] In addition, one objective of the present disclosure is to achieve intuitive user interaction by minimizing the gap between the high-dimensional vector space and the visualization space. means of solving the problem

[0011] According to one embodiment of the present disclosure, a computing device may be provided that includes a memory and at least one processor electronically connected to the memory, and is configured to execute at least one instruction stored in the memory, wherein the processor is configured to perform the operation of acquiring a first data set, the operation of selecting a first level among a plurality of levels classified according to a data processing method based on user input, the operation of determining at least one attribute of a first data processing model corresponding to the first level—the first data processing model includes at least one artificial intelligence model—the operation of providing prior information related to the task of processing the first data set using the first data processing model through a first GUI (Graphic User Interface), and the operation of constructing the first data processing model based on user input to the first GUI.

[0012] Additionally, according to one embodiment of the present disclosure, a data processing method may be provided configured to perform the following operations by at least one processor executing at least one instruction stored in memory: acquiring a first data set; selecting a first level among a plurality of levels classified according to a data processing method based on user input; determining at least one attribute of a first data processing model corresponding to the first level—the first data processing model includes at least one artificial intelligence model; providing prior information related to processing the first data set using the first data processing model through a first GUI (Graphic User Interface); and constructing the first data processing model based on user input to the first GUI.

[0013] Additionally, according to one embodiment of the present disclosure, an electronic device may be provided comprising a display implemented to display at least one GUI (Graphic User Interface), a memory, and at least one processor electronically connected to said memory, wherein at least one processor implemented to execute at least one instruction stored in said memory is configured to perform the following operations: receiving a user input selecting a first level among a plurality of levels classified according to a data processing method; displaying a first GUI (Graphic User Interface) through said display that represents prior information associated with processing said data set using a first data processing model corresponding to said first level—said that the first data processing model includes at least one artificial intelligence model—acquiring a diagnostic result for said data set based on a first vector set defined in a first-dimensional embedding area from said first data processing model; and providing said first vector set to a first visualization tool to display a first data image of two dimensions or three dimensions—said that the first data image corresponds to said first data set—through said display.

[0014] Additionally, according to one embodiment of the present disclosure, by at least one processor executing at least one instruction stored in memory, the operation of acquiring a data set; the operation of acquiring a vector set by embedding the data set into a specific dimension—the vector set includes a plurality of vectors corresponding to a plurality of data included in the data set—the operation of screening the vector set by calculating at least one feature value based on a vector included in the vector set using a screener having at least one metric set; the operation of identifying at least one vector in which the at least one feature value satisfies a predetermined condition; the operation of acquiring a data image including a plurality of data points representing the data set in two dimensions or three dimensions by processing the vector set using at least one visualization tool; the operation of determining a first region including at least one data point corresponding to the at least one vector on the data image and generating tag information associated with the first region; and a first scene including the first region—the first scene is generated by capturing the first region from a specific viewpoint—and snapshot information including the tag information. A method including a saving operation may be provided.

[0015] Additionally, according to one embodiment of the present disclosure, by at least one processor executing at least one instruction stored in the computer program, the operation of acquiring a data set; the operation of acquiring a vector set by embedding the data set into a specific dimension—the vector set includes a plurality of vectors corresponding to a plurality of data included in the data set—the operation of screening the vector set by calculating at least one feature value based on a vector included in the vector set using a screener having at least one metric set; the operation of identifying at least one vector in which the at least one feature value satisfies a predetermined condition; the operation of acquiring a data image including a plurality of data points representing the data set in two dimensions or three dimensions by processing the vector set using at least one visualization tool; the operation of determining a first region including at least one data point corresponding to the at least one vector on the data image and generating tag information associated with the first region; and a first scene including the first region—the first scene is generated by capturing the first region from a specific viewpoint—and a snapshot including the tag information. A computer program configured to perform the operation of storing information may be provided.

[0016] Additionally, according to one embodiment of the present disclosure, a computing device may be provided that includes a memory and at least one processor electronically connected to the memory, and is configured to execute at least one instruction stored in the memory, wherein the processor is configured to perform the operation of providing a first data image corresponding to a first data set—the first data image includes a plurality of data points corresponding to each of the data included in the first data set—through a first view port, the operation of generating first snapshot information including a first scene of the first data image being provided through the first view port in response to user input received through a first GUI and providing it through a second view port, and the operation of generating link information for connecting the first snapshot information to an external communication network based on user input to a second GUI provided through the second view port.

[0017] Additionally, according to one embodiment of the present disclosure, a data interaction method may be provided, comprising: an operation of providing a first data image corresponding to a first data set—the first data image includes a plurality of data points corresponding to each of the data included in the first data set—through a first view port by at least one processor executing at least one instruction in memory; an operation of generating first snapshot information including a first scene for the first data image being provided through the first view port in response to user input received through a first GUI and providing it through a second view port; and an operation of generating link information for connecting the first snapshot information to an external communication network based on user input for a second GUI provided through the second view port.

[0018] Additionally, according to one embodiment of the present disclosure, an electronic device may be provided that includes a display implemented to display at least one GUI (Graphic User Interface), a memory, and at least one processor electronically connected to said memory, wherein the at least one processor implemented to execute at least one instruction stored in said memory is configured to perform the following operations: providing a first view port through said display a first data image corresponding to a first data set—said that the first data image includes a plurality of data points corresponding to each of the data included in said first data set—in response to user input received through said GUI, providing a first snapshot information including a first scene of said data image being provided through said first view port through said display, in response to user input received through said GUI, and providing link information for connecting said first snapshot information to an external communication network based on user input to said GUI provided through said second view port.

[0019] Additionally, according to one embodiment of the present disclosure, a method may be provided comprising, by at least one processor executing at least one instruction stored in memory, the operation of visualizing a data set and providing a first data image corresponding to the data set; the operation of receiving user input for a first region on the first data image; the operation of determining a latent code corresponding to the first region and providing the determined latent code to a generation model to generate synthetic data; the operation of inputting the synthetic data into a data processing model—the data processing model is trained to embed data into a specific dimension—to obtain a synthetic vector corresponding to the synthetic data; and the operation of visualizing the synthetic vector and providing a synthetic point corresponding to the synthetic data on the first data image.

[0020] Additionally, according to one embodiment of the present disclosure, a computing device may be provided comprising a display implemented to display at least one GUI (Graphic User Interface), a memory, and at least one processor electronically connected to said memory, wherein at least one processor implemented to execute at least one instruction stored in said memory is configured to perform the following operations: visualizing a data set and providing a first data image corresponding to said data set through said display; receiving user input for a first area on said first data image; determining a latent code corresponding to said first area and providing the determined latent code to a generation model to generate synthetic data; inputting said synthetic data into a data processing model—said data processing model is trained to embed data into a specific dimension—to obtain a synthetic vector corresponding to said synthetic data; and visualizing said synthetic vector and providing a synthetic point corresponding to said synthetic data on said first data image through said display.

[0021] The means for solving the problem of the present invention are not limited to the means for solving the problem described above, and means for solving the problem not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from this specification and the attached drawings. Effects of the invention

[0022] According to one embodiment of the present disclosure, a computing device can automatically diagnose the quality of a large dataset and intuitively visualize it. This allows data analysts, AI model developers, or researchers to more easily understand the inherent structure of the data and formulate strategies to improve data quality.

[0023] In addition, according to one embodiment of the present disclosure, a user can explore and manipulate data more efficiently during the data visualization process and utilize it to improve the performance of a machine learning model.

[0024] The effects of the present disclosure are not limited to those described above, and unmentioned effects will be clearly understood by those skilled in the art from the present specification and the accompanying drawings. Brief explanation of the drawing

[0026] FIG. 1 is a diagram illustrating the configuration of a computing device according to various embodiments. FIG. 2 is a diagram illustrating various data processing methods included in a data clinic service provided by a computing device according to various embodiments. FIG. 3 is a diagram illustrating various systems for providing data clinic services and artificial intelligence models and algorithms for building such systems according to various embodiments. FIG. 4 is a diagram illustrating a method for a computing device to provide a data image according to various embodiments. FIG. 5 is a diagram illustrating a method for a computing device to acquire characteristics of a data set according to various embodiments. FIG. 6 is a drawing illustrating a data lens processing system and a data imaging system according to various embodiments. FIG. 7 is a diagram illustrating a system in which a computing device visualizes data and provides user interaction functions according to various embodiments. FIG. 8 is a diagram illustrating data visualization and interaction methods according to various embodiments. FIG. 9 is a diagram illustrating detailed steps performed in the imaging application step according to various embodiments. FIG. 10 is a diagram illustrating the results of data imaging and diagnosis according to the level of the lens and the visualization tool according to various embodiments. FIG. 11 is a drawing for illustrating a lens builder included in a computing device according to various embodiments. FIG. 12 is a diagram illustrating a method for a computing device to construct a data processing model for data imaging and diagnosis according to user input, according to various embodiments. FIG. 13 is a drawing for explaining information included in prior information according to various embodiments. FIG. 14 is a diagram illustrating an example of a computing device constructing a first data processing model according to various embodiments. FIG. 15 is a drawing illustrating another example in which a computing device constructs a first data processing model according to various embodiments. FIG. 16 is a drawing illustrating another example in which a computing device constructs a first data processing model according to various embodiments. FIG. 17 is a diagram illustrating a method for imaging and visualizing a data set using a first data processing model in which a computing device is constructed according to various embodiments. FIG. 18 is a diagram illustrating a method for a computing device to recommend a visualization tool according to various embodiments. FIG. 19 is a diagram illustrating a method for a computing device to generate synthetic data based on data imaging according to various embodiments. FIG. 20 is a diagram illustrating a method for a computing device to generate synthetic data based on user input according to various embodiments. FIG. 21 is a diagram illustrating an example of a method in which a computing device generates synthetic data based on user input according to various embodiments. FIG. 22 is a diagram illustrating a method for a computing device to remove data based on user input according to various embodiments. FIG. 23 is a diagram illustrating a method in which a computing device, according to various embodiments, performs data improvement in response to a data improvement request and provides a visual interaction therewith. FIG. 24 is a drawing illustrating an example of a computing device in which a data processing method including a snapshot function is implemented, according to various embodiments. FIG. 25 is a diagram illustrating a method for a computing device to screen a data set and generate snapshot information according to various embodiments. FIG. 26 is a diagram illustrating a method for a computing device to provide snapshot information based on user input according to various embodiments. Figure 27 is an example of a screen provided by a computing device. FIG. 28 is a diagram illustrating the function of a computing device to replay snapshot information according to various embodiments. FIG. 29 is a diagram illustrating a method in which a computing device generates diagnostic reports and snapshot information in conjunction according to various embodiments. Specific details for implementing the invention

[0027] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In describing the embodiments, technical details that are well known in the art to which the present disclosure belongs and are not directly related to the present disclosure will be omitted. This is intended to convey the essence of the present disclosure more clearly without obscuring it by omitting unnecessary explanations.

[0028] The embodiments described in this specification are intended to clearly explain the concept of the invention to those skilled in the art to which the invention pertains, and therefore the invention is not limited to the embodiments described in this specification, and the scope of the invention should be interpreted to include modifications or variations that do not deviate from the concept of the invention.

[0029] The terms used in this specification have been selected to be as widely used as possible, taking into account their functions in the present invention; however, these terms may vary depending on the intent of those skilled in the art to which the present invention pertains, case law, or the emergence of new technologies. However, if a specific term is defined and used with an arbitrary meaning, the meaning of that term will be described separately. Accordingly, the terms used in this specification should be interpreted based on their actual meaning and the content throughout this specification, rather than merely their names.

[0030] The drawings attached to this specification are intended to facilitate the explanation of the present invention. The shapes depicted in the drawings may be exaggerated as necessary to aid in understanding the present invention, and therefore the present invention is not limited by the drawings.

[0031] In this specification, each of the phrases such as “A or B”, “at least one of A and B”, “at least one of A or B”, “A, B or C”, “at least one of A, B and C”, and “at least one of A, B, or C” may include any one of the items listed together in the corresponding phrase, or all possible combinations thereof.

[0032] In cases where it is determined that a detailed description of known components or functions related to the present invention in this specification may obscure the essence of the invention, such detailed description shall be omitted as necessary. Furthermore, numbers used in the description of this specification (e.g., First, Second, etc.) are merely identification symbols to distinguish one component from another.

[0033] Furthermore, the suffixes "part" and "substance" used for components in the following description are assigned or used interchangeably solely for the ease of drafting the specification, and do not inherently possess distinct meanings or roles.

[0034] That is, the embodiments of the present disclosure are provided to make the present disclosure complete and to inform those skilled in the art of the scope of the present disclosure, and the invention of the present disclosure is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0035] Terms such as “first” and / or “second” may be used to describe various components, but said components shall not be limited by said terms. For the sole purpose of distinguishing one component from another, for example, without departing from the scope of rights according to the concept of the present disclosure, the first component may be named the second component, and similarly, the second component may be named the first component.

[0036] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. Conversely, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between. Other expressions describing the relationship between components, such as "between" and "exactly between," or "adjacent to" and "directly adjacent to," should be interpreted in the same way.

[0037] In the drawings, each block of the process flow diagrams and combinations of the flow diagrams can be executed by computer program instructions. Since these computer program instructions can be loaded into the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing equipment, the instructions executed through the processor of the computer or other programmable data processing equipment create means to perform the functions described in the flow diagram block(s). Since these computer program instructions can also be stored in computer-available or computer-readable memory that can be directed toward the computer or other programmable data processing equipment to implement the function in a specific way, the instructions stored in computer-available or computer-readable memory can also produce a manufactured item containing means of instruction to perform the function described in the flow diagram block(s). Since computer program instructions can be loaded onto a computer or other programmable data processing equipment, instructions that execute a computer or other programmable data processing equipment by performing a series of operation steps on the computer or other programmable data processing equipment to create a process executed by the computer may also provide steps for executing the functions described in the flowchart block(s).

[0038] Additionally, device-readable storage media may be provided in the form of non-transitory storage media. Here, 'non-transitory' simply means that the storage medium is a tangible device and does not contain a signal (e.g., electromagnetic waves), and this term does not distinguish between cases where data is stored semi-permanently and cases where it is stored temporarily.

[0039] Additionally, each block may represent a module, segment, or part of code containing one or more executable instructions for executing a specified logical function(s). It should also be noted that in some alternative execution examples, the functions mentioned in the blocks may occur out of order. For instance, two blocks described in succession may actually be executed substantially simultaneously, or the blocks may be executed in reverse order depending on the corresponding function. For instance, actions performed by a module, program, or other component may be executed sequentially, in parallel, iteratively, or heuristically; one or more of these actions may be executed in a different order, omitted, or one or more other actions may be added.

[0040] As used in this disclosure, the term “unit” refers to a software or hardware component such as a Field Programmable Gate Array (FPGA) or an Application Specific Integrated Circuit (ASIC). A “unit” performs specific roles but is not limited to software or hardware. A “unit” may be configured to reside in an addressable storage medium or may be configured to run one or more processors. Accordingly, according to some embodiments, a “unit” includes components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functions provided within the components and “units” may be combined into a smaller number of components and “units” or further separated into additional components and “units.” In addition, the components and 'parts' may be implemented to utilize one or more CPUs within the device or secure multimedia card. Furthermore, according to various embodiments of the present disclosure, the 'parts' may include one or more processors.

[0041] The operating principles of the present disclosure will be described in detail below with reference to the attached drawings. In describing the present disclosure below, specific descriptions of related known functions or configurations will be omitted if it is determined that such detailed descriptions would unnecessarily obscure the essence of the present disclosure. Furthermore, the terms described below are defined in consideration of their functions in the present disclosure, and these may vary depending on the intentions or practices of the user or operator. Therefore, their definitions should be based on the content throughout this specification.

[0043] FIG. 1 is a diagram illustrating the configuration of a computing device according to various embodiments.

[0044] Referring to FIG. 1, a computing device (100) according to one embodiment (an electronic device including computing means such as a server or client device, hereinafter referred to as "computing device") may include a processor (110), memory (120), storage device (130), communication circuit (140), and a bus (not shown). The configuration of the computing device (100) is not limited to the configuration shown in FIG. 1 or the configuration described above, and may further include hardware or software configurations included in general computing devices or mobile devices.

[0045] The processor (110) may include at least one processor implemented such that at least some parts provide different functions. For example, it may execute software (e.g., a program) to control at least one other component (e.g., a hardware or software component) of a computing device (100) connected to the processor (110) and perform various data processing or operations. According to one embodiment, as at least part of the data processing or operations, the processor (110) may store instructions or data received from other components in memory (120) (e.g., volatile memory), process the instructions or data stored in the volatile memory, and store the resulting data in non-volatile memory. According to one embodiment, the processor (110) may include a main processor (e.g., a central processing unit or an application processor) or an auxiliary processor (e.g., a graphics processing unit, a neural processing unit (NPU), an image signal processor, a sensor hub processor, or a communication processor) that can be operated independently or together with it. For example, if the computing device (100) includes a main processor and an auxiliary processor, the auxiliary processor may be configured to use less power than the main processor or to be specialized for a designated function. The auxiliary processor may be implemented separately from the main processor or as part thereof. The auxiliary processor may control at least some of the functions or states associated with at least one component of the computing device (100) (e.g., a display (240) or a communication circuit), for example, on behalf of the main processor while the main processor is in an inactive (e.g., sleep) state, or together with the main processor while the main processor is in an active (e.g., application execution) state.According to one embodiment, an auxiliary processor (e.g., an image signal processor or a communication processor) may be implemented as part of other functionally related components (e.g., a communication circuit). According to one embodiment, an auxiliary processor (e.g., a neural network processing unit) may include a hardware structure specialized for processing an artificial intelligence model. Meanwhile, the operation of the computing device (100) described below may be understood as the operation of the processor (110).

[0046] According to various embodiments, the memory (120) may include at least one memory implemented such that at least some parts provide different functions. The memory (120) may store various data used by at least one component of the computing device (100) (e.g., processor (110)). The data may include, for example, software (e.g., programs) and input data or output data for related instructions. The memory (120) may include volatile memory or non-volatile memory. The memory (120) may be implemented to store an operating system, middleware or applications, and / or the aforementioned artificial intelligence model.

[0047] Additionally, the memory (120) may include a plurality of instructions (121) that direct the operations of the processor (110) for implementing functions provided by the service. At this time, the processor (110) may execute at least some of the plurality of instructions stored in the memory (120). The computing device (110) may include a software server that includes a processor (110) that executes functions provided by the service based on at least some of the plurality of instructions.

[0048] The storage device (130) may provide a large-capacity storage device to the computing device (100). The storage device (130) may be a computer-readable medium. For example, the storage device (130) may be a floppy disk device, a hard disk device, an optical disk device, a tape device, a flash memory or other similar solid-state memory device, or a device array including a storage area network or other configuration devices. Additionally, a computer program product is explicitly implemented on an information medium. The computer program product contains instructions that perform one or more methods such as those described above upon execution. The information medium is a computer-readable medium or a machine-readable medium, such as memory (120), the storage device (130), or the memory of the processor (110).

[0049] Additionally, the storage device (130) may include a database (DB). The storage device (130) may include a database having a pre-structured data structure. The computing device (110) may store a data set having mutual associations in the database.

[0050] A computing device according to the present disclosure can provide services based on various artificial intelligence frameworks performed by at least one processor and a memory electronically connected to at least one processor.

[0051] In this regard, the memory (120) or storage device (130) may store at least one artificial intelligence model in which various types of artificial intelligence (or machine learning) frameworks are implemented that can be trained to perform a given task. For example, support vector machines, decision trees, neural networks, etc. are just a few examples of machine learning frameworks used in various applications such as image processing and natural language processing. Some artificial intelligence frameworks, such as neural networks, may utilize layers of nodes that perform specific operations.

[0052] In a neural network, nodes are connected to one another through one or more edges. A neural network may include an input layer, an output layer, and one or more intermediate layers. Individual nodes process their respective inputs according to a predefined function and provide outputs to subsequent layers or, in some cases, previous layers. The input to a specific node may be multiplied by a weight value corresponding to the edge between the input and the node. Additionally, nodes may have individual bias values ​​used to generate outputs. Various learning procedures can be applied to learn edge weights and / or bias values ​​(parameters).

[0053] A neural network structure may have multiple layers that perform different specific functions. For example, one or more node layers may collectively perform specific operations such as pooling, encoding, or convolution operations. In this disclosure, the term "layer" may refer to a group of nodes that share inputs and outputs, such as exchanging with an external source or other layers of the network. The term "calculation" may refer to a function that can be performed by one or more node layers. The term "model structure" may refer to the overall architecture of a layered model, including the number of layers, the connectivity of the layers, and the types of tasks performed by individual layers. The term "neural network structure" may refer to the model structure of a neural network. The terms "trained model" and / or "tuned model" may refer to the model structure along with parameters for the trained or tuned model structure. For example, two trained models can have different values ​​for parameters while sharing the same model structure, such as when two models are trained on different training data or when the training process involves an underlying probabilistic process.

[0054] "Transfer learning" is a broad approach to training a model with limited task-specific training data for a particular task. In transfer learning, a model is first pre-trained on another task where important training data is available, and then can be tuned to the specific task using task-specific training data.

[0055] As used in this disclosure, the term “pre-training” refers to training a model on a pre-training dataset to tune model parameters in a manner that allows for subsequent tuning of the corresponding model parameters to tune the model for one or more specific tasks. In some cases, pre-training may include a self-supervised learning process on unlabeled training data, wherein the “self-supervised” learning process includes learning from the structure of pre-training examples in the absence of explicit (e.g., manually provided) labels. Subsequent modification of the model parameters obtained through pre-training is referred to herein as “tuning.” Tuning may be performed on one or more tasks using supervised learning on explicitly labeled training data, and in some cases, a different task from pre-training may be used for tuning.

[0056] A communication bus (not shown) may be a component for electronically (or communically) connecting multiple components included in a computing device. That is, each component may be interconnected using various buses and mounted on a common motherboard or mounted in other appropriate ways.

[0057] The input / output interface (not shown) may include an input interface connected to an input device to receive an input signal, or an output interface connected to an output device to output an output signal.

[0058] Additionally, the computing device (100) may further include at least one communication circuit (140) for communicating with an external device.

[0059] The communication circuit (140) can support the establishment of a direct (e.g., wired) communication channel or a wireless communication channel between the computing device (100) and an external computing device, and the performance of communication through the established communication channel. The communication circuit may include one or more communication processors (e.g., communication chips) that operate independently of the processor (110) (e.g., program processor) and support direct (e.g., wired) communication or wireless communication. According to one embodiment, the communication circuit (140) may include a wireless communication module (e.g., cellular communication module, short-range wireless communication module, or GNSS (global navigation satellite system) communication module) or a wired communication module (e.g., LAN (local area network) communication module, or power line communication module). Among these communication modules, the corresponding communication module can communicate with an external computing device through a first network (e.g., a short-range communication network such as Bluetooth, WiFi (wireless fidelity) direct, or IrDA (infrared data association)) or a second network (e.g., a legacy cellular network, a 5G network, a next-generation communication network, the Internet, or a computer network (e.g., a LAN or WAN)). These various types of communication modules may be integrated into a single component (e.g., a single chip) or implemented as multiple separate components (e.g., multiple chips). The wireless communication module can identify or authenticate the computing device (100) within a communication network, such as the first network or the second network, using subscriber information (e.g., International Mobile Subscriber Identifier (IMSI)) stored in a subscriber identification module. The wireless communication module can support 5G networks following 4G networks and next-generation communication technologies, such as NR access technology (new radio access technology).NR access technology can support high-speed transmission of high-capacity data (enhanced mobile broadband (eMBB)), minimization of terminal power and connection of multiple terminals (massive machine type communications (mMTC)), or high reliability and low latency (ultra-reliable and low-latency communications (URLLC)). The wireless communication module can support a high-frequency band (e.g., mmWave band) to achieve a high data transmission rate, for example. The wireless communication module can support various technologies to secure performance in the high-frequency band, for example, beamforming, massive MIMO (multiple-input and multiple-output), full-dimensional MIMO (FD-MIMO), array antenna, analog beamforming, or large-scale antenna. The wireless communication module can support various requirements specified in the computing device (100), endoscope device, or network system. According to one embodiment, the wireless communication module may support a Peak data rate (e.g., 20 Gbps or higher) for realizing eMBB, loss coverage (e.g., 164 dB or lower) for realizing mMTC, or U-plane latency (e.g., downlink (DL) and uplink (UL) each of 0.5 ms or lower, or round trip 1 ms or lower) for realizing URLLC.

[0060] The computing device (100) may be implemented to include at least some of the above-described configurations (processor, communication circuit, memory, display). For example, a user device may be implemented to include a processor, a communication circuit, memory, a sensor, and a display. Additionally, for example, a server device may be implemented to include a processor, a communication circuit, and memory.

[0062] FIG. 2 is a diagram illustrating various data processing methods included in a data clinic service provided by a computing device according to various embodiments.

[0063] Referring to FIG. 2, the data clinic service may include various data processing methods. At this time, the various data processing methods may be coded and stored in the memory of the computing device, and at least one processor included in the computing device may be configured to execute at least one coded instruction. Specifically, at least one processor may process a received input data set based on the various data processing methods and output an output data set.

[0064] For example, a computing device according to various embodiments of the present disclosure may perform a method for data imaging, a method for data enhancement, a method for data generation, a method for extracting data features, or a method for data evaluation, but is not limited thereto.

[0065] In addition, each of the above-described operation methods can be performed based on operation algorithms of at least one processor included in the computing device.

[0066] For example, a computing device according to various embodiments of the present disclosure may perform a data imaging algorithm, a data enhancement algorithm, a data generation algorithm, a data feature extraction algorithm, or a data evaluation algorithm, but is not limited thereto.

[0067] In this case, since the names of each operation method and algorithm are arbitrarily named according to the output results for the convenience of explanation, each operation method or algorithm is defined merely based on the operations performed by the processor and does not limit the invention by the name of the operation method or algorithm itself.

[0068] More specifically, a computing device according to various embodiments of the present disclosure can process an input data set according to a data imaging algorithm to generate an image of the input data set.

[0069] In addition, a computing device according to various embodiments of the present disclosure can improve data by processing an input data set according to a data improvement algorithm and can generate the result of the improvement.

[0070] In addition, a computing device according to various embodiments of the present disclosure can generate synthetic data by processing an input data set according to a data generation algorithm.

[0071] In addition, a computing device according to various embodiments of the present disclosure can process an input data set according to a data feature extraction algorithm to extract a property of the input data set.

[0072] In addition, a computing device according to various embodiments of the present disclosure can process an input data set according to a data evaluation algorithm to evaluate the quality of the input data set.

[0073] Detailed information regarding each of the aforementioned algorithms is explained below.

[0074] In addition, a computing device according to various embodiments of the present disclosure may perform the various operation methods or algorithms described above in parallel, sequentially, or selectively. Specifically, the computing device may use the same input data in parallel as input values ​​for different algorithms, may sequentially use a result value output according to a specific algorithm as an input value for another algorithm, or may selectively perform some of the algorithms among a plurality of algorithms according to a predetermined method.

[0075] In addition, various operation methods or algorithms for the data clinic described above may be performed in a deep learning model included in a computing device according to various embodiments of the present disclosure. Specifically, a computing device according to various embodiments of the present disclosure may include one deep learning model for performing the various operation methods or algorithms described above, but is not limited thereto, and may include a plurality of deep learning models for performing each of the operation methods or algorithms described above, or may include one or more deep learning models for performing at least some of the various operation methods or algorithms described above.

[0077] FIG. 3 is a diagram illustrating various systems for providing data clinic services and artificial intelligence models and algorithms for constructing such systems according to various embodiments. Here, a system may refer to a system comprising at least one software configuration or hardware configuration to perform a specific function.

[0078] Referring to FIG. 3, a computing device according to the present disclosure may include a data clinic system composed of various artificial intelligence (or neural network, machine learning, etc.) models to provide clinic services.

[0079] For example, computing devices may include, but are not limited to, data imaging systems, data diagnostic systems, and data treatment systems.

[0080] Here, the data imaging system may include, but is not limited to, a lens processing model for determining the optimal dimension for representing the characteristics of the data, an imaging model for acquiring a data image that reflects the intrinsic characteristics of the data, or a visualization model for visually representing the data.

[0081] In addition, the data diagnostic system may include, but is not limited to, a diagnostic model for diagnosing at least one characteristic of the data or a quality evaluation model for evaluating the quality of the data.

[0082] Additionally, the data treatment system may include, but is not limited to, a synthetic model (or generative model) for generating targeted virtual data (or synthetic data) as needed, a data diet model for removing at least a portion of the data, or a data correction model for adjusting the characteristics of at least a portion of the data.

[0083] Various machine learning models included in a computing device may be composed of multiple modules stored in memory. In this disclosure, a module may include multiple hardware components for implementing an artificial intelligence model that performs a specific function. For example, a module may include, but is not limited to, an encoder, a decoder, a generator, a discriminator, an adapter, a natural language processing module, or a large language model (LLM).

[0084] A computing device can store the aforementioned multiple modules and can obtain an artificial intelligence model for data clinic by configuring an artificial intelligence framework based on at least some of the multiple modules. For example, a data lens included in a data imaging system may be implemented as an artificial intelligence model including at least one encoder or at least one adapter, but is not limited thereto.

[0086] FIG. 4 is a diagram illustrating a method for a computing device to provide a data image according to various embodiments.

[0087] Referring to FIG. 4, a computing device according to various embodiments of the present disclosure can receive a data set and provide an Image of Data (IOD).

[0088] At this time, the data set may be M (M>0) dimensional data. In other words, the data set may be a data set defined on an M-dimensional input space (310).

[0089] Additionally, the data set may be a data set of a single modality. For example, the data set may be an image data set. Additionally, the data set may be a text data set. Additionally, not limited thereto, the data set may be a set of data with different modalities. For example, the data set may be an image data set containing annotation information. Additionally, the data set may be a mixed data set of images and text.

[0090] A computing device according to various embodiments of the present disclosure can receive and process data of all modalities that can be used for deep learning training, such as time series data sets and sensor data sets, as input data sets, as well as the image data and text data described above.

[0091] The data image (IOD) provided by the computing device according to various embodiments of the present disclosure may be displayed in an imaging space (320) by processing an input data set. Here, the image does not refer to a 2D image, but is a general term for a visual representation of data. Specifically, the imaging space (320) is a concept that includes a 2D space, a 3D space, and an N-dimensional virtual space, and refers to a space where the data image provided according to the embodiment appears. For example, when the computing device processes an input data set and outputs the data image in PDF form, it may output an output in which the data image is displayed in a 2D or 3D imaging space, but is not limited thereto.

[0092] In cases where a computing device according to various embodiments of the present disclosure includes an output device (not shown), the computing device may provide a data image through the output device. For example, the computing device may provide a data image by outputting the data image through a display. In this case, the imaging space (320) may be the screen of the display. Additionally, for example, the computing device may provide a data image by outputting the data image through a printing device. In this case, the imaging space (320) may be paper output by the printing device.

[0093] Additionally, when a computing device according to various embodiments of the present disclosure communicates with an external device through a communication unit, the computing device may provide a data image through the external device. In this case, the imaging space (320) may be a display screen of the external device. For example, if the computing device is a server device, the server device may provide a data image by transmitting the data image to at least one external device that communicates with the server device through a network connected to the server device.

[0094] A computing device according to various embodiments of the present disclosure can acquire a data image based on a vector set (or data point set, point data set, etc.) (330) corresponding to an input data set.

[0095] At this time, the computing device can obtain a vector set by mapping the data included in the input data set into a specific dimensional embedding space (or latent space). Specifically, the computing device can obtain a vector set by identifying the manifold formed by the data set in the specific dimensional embedding space. Here, the manifold may refer to the shape represented by the input data set in the specific dimensional embedding space. In other words, the manifold may refer to the region where the vector set is identified or the shape formed by the vector set when the input data set is mapped into a vector set in the specific dimensional embedding space.

[0096] A data image (IOD) may be a data set in which each data included in the data set is visualized as points. In this case, since the shape or color of the visualized points can be selected in various ways depending on the embodiment, the invention is not intended to be limited by the term 'point' itself. Furthermore, points may be expressed in various terms depending on the embodiment. For example, points may be expressed in terms such as vectors or features appearing in an embedding space or latent space, but are not limited thereto.

[0097] In order for a computing device according to various embodiments of the present disclosure to provide a data image, it is necessary to identify a vector set corresponding to an input data set, as described above.

[0098] A computing device can obtain a vector set by mapping a dataset into an N-dimensional embedding space based on a predetermined condition defined by a mapping function (e.g., a matrix stored in advance to map to an embedding space of a specific dimension). For example, the computing device can obtain a vector set by encoding the dataset, but is not limited thereto. For example, the computing device can obtain a vector set by inputting the dataset into a pre-trained encoder and through the output layer of the encoder, but is not limited thereto.

[0099] Here, embedding refers to the process of converting high-dimensional data into low-dimensional vectors to maintain similarities and structural relationships between data points. This embedding is performed in a manner that preserves the core information of the data while increasing computational efficiency. During the embedding process, each data point is represented as a vector, and these vectors can reflect the distribution and characteristics of the entire dataset. This lays the foundation for analyzing the statistical characteristics and inherent patterns of the dataset.

[0100] The computing device may include a data lens (400) for acquiring a data image (IOD) by embedding and visualizing a data set. In this case, the data lens (400) may include at least one processing configuration for processing data. Specifically, the data lens (400) may include at least one neural network model (e.g., encoder, etc.) for acquiring a vector set based on the data set and at least one visualization model (e.g., PCA, T-SNE, or UMAP, etc.) for acquiring a data image by visualizing the data set based on the vector set. Specifically, the data lens (400) may acquire a vector set corresponding to the data set by embedding the data set in an N-dimensional latent space and may acquire a data image (IOD) corresponding to the data set by displaying the vector set in an M-dimensional (e.g., 2-dimensional or 3-dimensional) imaging space (320).

[0102] FIG. 5 is a diagram illustrating a method for a computing device to acquire characteristics of a data set according to various embodiments.

[0103] Referring to FIG. 5, the computing device can process the acquired data set to obtain characteristic information corresponding to the data set.

[0104] A property of a data set or data may include information associated with the distribution of the data set or data (e.g., geometric distribution or statistical distribution, etc.). Specifically, the property information may include the property values ​​of each data included in the data set. For example, a computing device may obtain property information representing the distribution of the property values ​​of the data included in the data set. Additionally, the computing device may obtain property information of the data set based on statistical distributions, such as the mean, deviation, or variance of the property values ​​of each data.

[0105] For example, a property of a dataset or data may include intrinsic properties associated with the distribution of the dataset itself. For example, a property of a dataset or data may include, but is not limited to, the density, homogeneity, bias, or distribution of the dataset or data.

[0106] As another example, a property of a dataset or data may include a task-dependent property related to the task in which the dataset is utilized (e.g., Classification). For example, a property of a dataset or data may include, but is not limited to, a labeling error rate or a ratio of geometrically adjacent data pairs of different classes (hard-negative).

[0107] Additionally, the computing device may store computational metrics corresponding to each of the characteristics of the data set or data in memory. More specifically, the computing device may store metrics for calculating the density of the data set or data, metrics for calculating the homogeneity of the data set or data, metrics for calculating the bias of the data set or data, or metrics for calculating the distribution of the data set or data, but is not limited thereto.

[0108] In addition, the computing device can acquire features of a data set based on stored computational metrics according to a data feature extraction algorithm built with an artificial neural network. Specifically, the feature extraction algorithm can be implemented as a feed-forward neural network.

[0109] For example, a computing device may include a separate neural network for computing features of a data set, or a neural network including a layer for computing features of a data set, but is not limited thereto.

[0110] For example, a computing device may include an artificial neural network for feature extraction designed to extract features of a data set when it receives a data set as input. In this case, the artificial neural network for feature extraction may be an artificial neural network that has been transferred to compute features of the data.

[0111] As another example, a computing device can acquire features of a data set by constructing an artificial neural network that adds a layer for extracting data features to a neural network model (e.g., data lens, encoder, etc.) for providing data images based on a data set. Specifically, the computing device can identify a set of vectors based on a data set and acquire features of the data set or data based on the identified set of vectors.

[0112] At this time, the computing device can obtain characteristic values ​​for each data included in the data set by processing each vector included in the vector set using a predetermined algorithm. In this case, the computing device can calculate characteristic values ​​based on the geometric or statistical distribution of each vector included in the vector set and assign the calculated characteristic values ​​to the corresponding data. At this time, the characteristic values ​​can be calculated based on the distance between the vectors. For example, the characteristic values ​​can be obtained based on the number of vectors existing within a predetermined distance centered on a specific vector (or data point), but are not limited thereto. Additionally, for example, the characteristic values ​​can be obtained based on the average value of the distances to a predetermined number of vectors close to a specific vector, but are not limited thereto. For example, the computing device can calculate an average distance value based on the distance values ​​to K vectors close to a specific vector, and can obtain a first characteristic value (e.g., density, etc.) of the specific vector based on the calculated average distance value, but is not limited thereto.

[0114] To optimize the framework of an artificial intelligence model and derive high-accuracy results, the quality of the data used for model training is critical.

[0115] As previously mentioned, since data quality is a concept that encompasses both quantitative and qualitative quality, for the successful training of an artificial intelligence model, it is necessary to (i) secure a sufficient amount of training data to train the model, (ii) secure training data with high-quality intrinsic characteristics (e.g., unbiased distribution), and (iii) secure training data with characteristics (e.g., task-dependent property) suitable for the training objective (e.g., the task of the artificial intelligence model).

[0116] A computing device according to one embodiment of the present disclosure may synthesize, modify (or adjust), or remove data in a way that enhances the intrinsic characteristics and task-dependent characteristics of the data set in order to obtain high-quality training data.

[0117] In addition, the computing device according to the present disclosure can improve the overall quality of a data set by removing at least some of the data from the data set.

[0118] Generally, data downsampling or undersampling techniques are used to address data imbalance. However, existing undersampling methods have a problem of adversely affecting the training of machine learning models because they remove data without considering the characteristics of the machine learning dataset.

[0119] A computing device according to the present disclosure can improve the learning efficiency of an artificial intelligence model being learned by appropriately removing at least some data from a data set.

[0121] FIG. 6 is a drawing illustrating a data lens processing system and a data imaging system according to various embodiments.

[0122] Referring to FIG. 6(a), a computing device may acquire a System of Lens based on a dataset. In this case, the term System of Lens may refer to at least one configuration for mapping the dataset into a specific embedding space. For example, the System of Lens may include at least one encoder for mapping the dataset into an embedding space (or latent space) of a specific dimension and / or at least one adapter for adjusting parameters. Additionally, for example, the System of Lens may include a neural network layer composed of at least one node for identifying a latent variable (or latent feature vector) corresponding to the dataset.

[0123] The computing device (3500) can determine a data lens system that processes the data set to preserve the intrinsic characteristics of the data set based on the data set.

[0124] For example, a computing device can obtain a lens system corresponding to a data set based on a database. Specifically, the computing device can search for a lens system corresponding to the data set in the database based on the characteristics of the input data set.

[0125] As another example, a computing device can obtain a lens system corresponding to a data set based on a lens processing algorithm. Specifically, the computing device can compute the optimal dimensionality that preserves the intrinsic characteristics of the input data set.

[0126] Referring to FIG. 6(b), a computing device can obtain an image of data by processing a data set based on a data imaging system (3510) that includes a determined data lens system. At this time, the imaging system (3510) may include a lens system composed of at least one module (e.g., encoder, adapter, etc.).

[0127] In this case, the computing device can acquire a data image representing the intrinsic characteristics of the data set using an imaging system (3510).

[0129] The present disclosure provides a method for vectorizing data to precisely identify its intrinsic distribution and intuitively visualizing it in a two-dimensional or three-dimensional space, in order to effectively analyze and visually represent a large dataset. In particular, by applying interaction and UI / UX implementation techniques to minimize the gap between the high-dimensional vector space and the visualization space, the user can more easily understand and manipulate the data quality.

[0130] In addition, the method of the present disclosure can receive a large data set as input and provide a clearer visual representation while maintaining the intrinsic characteristics of the data set (e.g., geometric distribution or relationships between data). Through this, users can intuitively grasp the distribution of the data and effectively utilize it to improve data quality during the learning of artificial intelligence or big data analysis process.

[0132] FIG. 7 is a diagram illustrating a system in which a computing device visualizes data and provides user interaction functions according to various embodiments.

[0133] Referring to FIG. 7, a computing device (700) can image a data set (710) to obtain a vector set corresponding to the data set (710). At this time, the computing device (700) can obtain a data image (720) based on the vector set. The data image (720) may be data visualized in two dimensions or three dimensions from the data set (710). The computing device (100) can identify an N (N>3) dimensional vector set based on the data set (710) and obtain a data image (720) by reducing the dimensionality of the N dimensional vector set to two dimensions or three dimensions. That is, the computing device (700) can represent the data image (720) corresponding to the data set (700) in a two-dimensional imaging space or a three-dimensional imaging space.

[0134] A computing device (700) can provide a data image (720) through a network environment (e.g., a web or app environment). A user device (701) can view the data image (720) through the network environment and obtain information about the viewed data image (720). At this time, the user device (701) may be an electronic device such as a laptop computer, a smartphone, a tablet, etc. Additionally, for example, the user device (701) may be a wearable device such as a smart watch, smart glasses, smart eyeglasses, etc. Additionally, for example, the user device (701) may be a media device such as a streaming media device, a media player, an automotive entertainment system, etc. Additionally, for example, the user device (701) may be an XR (mixed reality) device such as a VR device or a device including AR glasses, etc. That is, the user device (701) may be an electronic device that connects to a network to receive data visualization services according to various embodiments of the present disclosure.

[0136] FIG. 8 is a diagram illustrating data visualization and interaction methods according to various embodiments.

[0137] Referring to FIG. 8, the data visualization and interaction method may include a data imaging application step (S810), a data imaging and diagnosis step (S820), a data visualization step (S830), and an interaction step (S840).

[0138] In the data imaging request step (S810), the electronic device may request imaging, diagnosis, or visualization of a data set from the server. Specifically, the electronic device may transmit the data set and request information for the data set to the server. The server may process the data set based on the received data set and request information.

[0139] In the data imaging and diagnosis step (S820), the server can image the data set by vectorizing the data set. Specifically, the server can identify a vector set corresponding to the data set by embedding the data set into a latent space of a specific dimension.

[0140] The server can diagnose the intrinsic properties and task-dependent attributes of the dataset by analyzing the distance, neighbor relationship, or cluster structure between individual data points included in the dataset based on the identified vector set.

[0141] For example, to measure data density as an intrinsic characteristic, the server may analyze the distance distribution to K neighbor vectors close to a specific data point, or identify areas that are excessively densely packed into a specific class to determine bias. Additionally, to evaluate homogeneity, it may measure the diversity or degree of dispersion of the regions to which each vector belongs.

[0142] In addition, for example, the server can diagnose task-dependent characteristics by identifying how each class is distributed in vector space (such as boundaries or overlaps between classes) for a dataset used in a classification task. Through this, the server can detect cases where there are imbalanced classes or where there are many points (hard negatives) where a specific class is excessively close to another class.

[0143] In this process, the server can detect outliers or identify factors that degrade data quality (e.g., incorrect labels, duplicate data, noisy samples, etc.) by considering at least one of the Euclidean distance, cosine similarity, or other distances between vectors. Additionally, the server can pass these analysis results to subsequent steps (e.g., data cleaning or label correction) to be used for improving the quality of the dataset.

[0144] In the data visualization step (S830), the server can visualize and represent the data set in a two-dimensional or three-dimensional space. The server can obtain a data image (IOD) by dimensionally reducing the data set or a vector set corresponding to the data set, and provide the data image in a two-dimensional or three-dimensional space. For example, the server can apply dimensionality reduction techniques such as PCA (Principal Component Analysis), t-SNE, and UMAP, or use a neural network-based embedding model to map high-dimensional data to a low-dimensional visualization space.

[0145] Here, the server can provide visualized results reflecting the diagnostic results (cluster structure, outliers, bias, imbalanced classes, etc.) derived from the previous step (e.g., data diagnosis step). Specifically, the server can guide users to grasp data distribution and quality issues at a glance by marking data points identified as outliers with separate colors or special markers (e.g., triangles, stars, etc.), or by visually highlighting areas with ambiguous class boundaries or high density (e.g., displaying area borders, gradients).

[0146] In the interaction step (S840), the server can process inputs associated with the visualized data image received from an electronic device (e.g., user device) to provide a dynamic response to the visualized data image. Specifically, the user can perform inputs such as touch, mouse click, drag, or pinch zoom on any point within the visualized data image (e.g., a specific data point, cluster, or selected area).

[0147] For example, when a user clicks on a data point, the computing device (server) of the present disclosure may provide a response by querying metadata associated with the data point (e.g., individual feature values, label information, statistical values, etc.) and transmitting the information to the user device, which can be displayed in a pop-up window, tooltip, or on a separate layer. As another example, when a user drags a specific section in a visualized space to designate an area, the server may further analyze the statistical distribution, density, class distribution, etc. of the data points within the designated area and provide the results (e.g., mean value, deviation, representative image, sample size, etc.) to the user device.

[0148] In addition, users can observe the data in various ways by performing interactions such as scaling down, enlarging, or rotating the visualized data image (in the case of 3D visualization), or changing the color scale. In response to these change requests, the server generates an updated data image by reapplying dimensionality reduction techniques (PCA, t-SNE, UMAP, etc.) or color and shape mapping algorithms, or by adjusting parameters, and provides it to the user device.

[0149] Thus, through the interaction step (S840) of the present disclosure, the user can intuitively and immediately explore the visualized data, gain a deeper understanding of the data's inherent structure, or detect potential quality issues (e.g., label errors, outliers, unclear boundaries between clusters, etc.) at an early stage. Furthermore, this interaction function can be effectively utilized by machine learning model developers during the process of performing data cleaning or label correction tasks.

[0150] The various interaction functions described in this step are merely examples to clarify the embodiments, and it is obvious to those skilled in the art that methods of responding to other forms of user input not described above (e.g., gesture input, voice, text commands, etc.) can be modified and applied. Therefore, it should be understood that the scope of the present invention is not limited by the specific examples used in this specification.

[0151] FIG. 9 is a diagram illustrating detailed steps performed in the imaging application step according to various embodiments.

[0152] During the process of applying for data imaging and diagnosis, the user can input settings for tools to be used for data imaging and diagnosis (e.g., AI models for data imaging of lenses, etc.), data visualization tools, etc.

[0153] Referring to FIG. 9, the data imaging application step (S810) may include a lens level selection step (S811), a lens attribute determination step (S813), other setting steps (S815), and a visualization tool selection step (S817).

[0154] In the lens level selection step (S811), the user may select a level of lens that includes at least one artificial intelligence model to be used for data imaging. In the present disclosure, the level of the lens may be classified and defined according to pre-set criteria related to the data imaging method. For example, the level of the lens may include, but is not limited to, a first level that diagnoses only quantitative indicators of a data set (basic statistics, missing value ratio, data outlier detection, etc.), a second level that diagnoses by vectorizing the data set using a pre-stored (pre-trained) artificial intelligence model, or a third level that diagnoses by vectorizing an artificial intelligence model optimized for the data set after additional training (fine-tuning) or new training. For example, if the scale of the data set is very large or high precision is required, a more sophisticated diagnostic result can be obtained by selecting the third level and directly retraining the model.

[0155] In the lens attribute determination step (S813), the user can determine the attributes of the lens, including at least one artificial intelligence model to be used for data imaging. In the present disclosure, the attributes of the lens may include the structure of the artificial intelligence model, hyperparameters (e.g., number of layers, number of parameters, learning rate, batch size, etc.), or search strategies (e.g., type of optimizer, initialization method), etc. For example, the user can balance analysis and processing time by adjusting the "model complexity (number of parameters)" or setting the "number of training epochs." Additionally, the user can choose whether to use a pre-trained model suitable for a specific domain (image, text, structured data, etc.) or to use a generic AI model.

[0156] In the other configuration step (S815), detailed options may be provided to reflect user requirements. For example, the user device may configure whether to perform data diagnosis by selecting whether to skip the diagnosis process and perform only simple visualization, or to perform diagnosis as well. Additionally, for example, the user device may configure whether to visualize diagnosis results by deciding whether to display cluster density or bias indicators on a separate color scale. Furthermore, for example, the user device may configure community creation criteria related to the method of creating subgroups or communities based on similarity (distance) criteria between data points or specific characteristics (label information, etc.). Additionally, for example, the user device may configure whether to generate a snapshot of the data distribution (visualization state) at a specific point in time (view-point) separately for comparison or analysis.

[0157] In the visualization tool selection step (S817), the user may select at least one of a plurality of pre-linked visualization tools (e.g., PCT, UMAP, T-SNE, etc.). In this case, the user may make a selection by considering the characteristics of each tool (dimension reduction method, ease of interpreting results, processing speed, etc.). Additionally, by selecting a pre-set visualization dimension (e.g., 2D, 3D), the user may decide whether to view it in a simple plane (projection) or to interact with it in 3D space, such as rotation or scaling.

[0158] For example, PCA is fast and intuitive to interpret, whereas t-SNE and UMAP have the advantage of reflecting the data cluster structure more precisely. Users can select a visualization tool by comprehensively considering the characteristics of the dataset, analysis goals, computing resources, etc.

[0159] As described above, in the data imaging application step (S810), the user can directly input various settings and decisions to control the data imaging and diagnosis process so that it is optimized for their analysis objectives and environment. This user-centric setting step supports the more effective execution of vectorization, diagnosis, and visualization processes to be performed in subsequent steps (S820, S830, etc.), and ultimately contributes to improving data quality and enhancing the performance of artificial intelligence models.

[0161] FIG. 10 is a diagram illustrating the results of data imaging and diagnosis according to the level of the lens and the visualization tool according to various embodiments.

[0162] FIG. 11 is a drawing for illustrating a lens builder included in a computing device according to various embodiments.

[0163] The computing device can construct a lens for imaging a data set according to settings entered by the user, and can diagnose and visualize the data set using the constructed lens and visualization tools.

[0164] A computing device can construct a lens for imaging a dataset according to settings entered by a user, and can diagnose and visualize the dataset using the constructed lens and visualization tools. Referring to FIG. 10, a computing device (e.g., a server) can provide a dataset to a lens builder. Here, the lens builder is a component for constructing an appropriate lens (e.g., a tool set for data imaging or diagnosis) by configuring or combining, or training or tuning, tools for processing the dataset (e.g., an artificial intelligence model, a statistical calculator, etc.) that reflect user settings for imaging and diagnosing the dataset. The lens builder may be implemented as a separate, independent hardware device or software module, or it may be implemented in a form where at least one processor executes multiple instructions stored in memory for lens construction.

[0165] For example, referring to FIG. 11, the lens builder may include, but is not limited to, a model storage unit including a plurality of artificial intelligence models, a model learning unit for learning the artificial intelligence models, an attribute determining unit for determining the attributes of the artificial intelligence models, a computation tool storage unit including a plurality of computation units for computing the statistical characteristics of the data, and a lens storage unit for storing the constructed lenses.

[0166] Specifically, the model storage unit may include, but is not limited to, a first storage unit storing multiple pre-trained models (e.g., first pre-trained model, second pre-trained model, etc.), a second storage unit storing multiple foundation models (e.g., first foundation model, second foundation model, etc.), and a third storage unit storing multiple adapters (e.g., first adapter, second adapter, etc.). Here, an adapter is a component electronically connected to an artificial intelligence model to convert an existing artificial intelligence model, and may be referred to as a data converter, an artificial intelligence converter, a low-rank adapter, or a conversion module. Additionally, the computational tool storage unit may store multiple computational units (e.g., first computational unit, second computational unit, etc.), and the lens storage unit may store multiple lenses (e.g., first lens, second lens, third lens, fourth lens, etc.), but is not limited thereto.

[0167] Referring again to FIG. 10, a computing device can construct at least one lens for imaging or diagnosing a data set using a lens builder. At this time, the at least one lens may include a computing tool that stores at least one instruction for processing the data set and producing a characteristic associated with the data set.

[0168] For example, a computing device can diagnose a data set by statistically analyzing the data set.

[0169] Specifically, the computing device may generate a first lens (LENS #1) based on a request for a first-level lens. The first lens may include at least one computational tool for diagnosing the statistical characteristics of a data set.

[0170] For example, referring to FIGS. 10 and 11, a computing device can use a lens builder to create a first lens comprising at least one of a plurality of computing units stored in a computing tool storage unit.

[0171] For example, the computing device may configure a first lens by selecting or combining at least one of a plurality of computing units (e.g., a missing value detector, an outlier detector, a variance / deviation calculator, a correlation analyzer, etc.) stored in a computing tool storage unit. For example, if a user requests "basic statistical level diagnosis," the computing device activates a computing unit that calculates the mean, minimum, maximum, or standard deviation, and if "missing value ratio diagnosis" is additionally selected, the first lens may be constructed to include a missing value detector.

[0172] In addition, the computing device inputs a data set into the first lens, and through the first lens, a first diagnostic result (1) for the data set st A DIAGNOSIS RESULT can be obtained. The first diagnostic result may include, for example, basic indicators indicating the state of data quality such as the total number of data, the ratio of missing values, and the ratio of outliers; statistical values ​​for specific attributes (mean, variance, skewness, kurtosis, etc.); results of analysis of correlations (Pearson correlation coefficient, Spearman correlation coefficient, etc.); or data counts by class, bias measures, etc.

[0173] This allows users to quickly grasp the overall statistical characteristics of the data set and decide on subsequent tasks such as handling missing values ​​or removing outliers.

[0174] As another example, computing devices can perform more in-depth characteristic diagnosis by vectorizing a data set and analyzing the vectorized data.

[0175] Specifically, the computing device can generate a second lens (LENS #2) using an appropriate pre-trained AI model based on a request for a second-level lens.

[0176] For example, referring to FIGS. 10 and 11, the computing device can generate a second lens by loading one or more models that match user requirements or data types (image, text, structured data, etc.) from among a plurality of pre-trained models (e.g., CNN models for computer vision, Transformer models for text embeddings, etc.) stored in the first storage unit of the model storage unit.

[0177] In addition, the computing device inputs a data set into at least one pre-trained model, and from at least one layer of the artificial intelligence model, a first vector set (1) corresponding to the data set st A VECTOR SET) can be obtained. The computing device calculates at least one characteristic value of the data included in the data set based on the first vector set, thereby obtaining a second diagnostic result (2 nd A DIAGNOSIS RESULT can be obtained. That is, the computing device can analyze the first set of vectors (e.g., Euclidean distance, clustering, density calculation, etc.) to produce a second DIAGNOSIS RESULT that provides higher-level insights into the intrinsic characteristics of the dataset, class distribution, bias, etc.

[0178] The second diagnostic result may include, for example, but is not limited to, cluster structures between data in the embedding space (number of clusters, representative centroids), density / variance between vectors (whether data is excessively concentrated in a specific area), boundary ambiguity by class (hard negative ratio, distance distribution between classes, etc.), or differences between subgroups based on specific attribute values.

[0179] Through this, unlike simple statistical characteristics, it is possible to capture the inherent structure of the data or potential bias issues, and it is easy to identify the causes of errors or performance degradation that may occur during model training.

[0180] In addition, the computing device can generate a third lens (LENS #3) to perform vectorization by training an artificial intelligence model optimized for the dataset based on a request for a third-level lens.

[0181] For example, referring to FIGS. 10 and 11, a computing device may generate a third lens comprising at least one of a plurality of foundation models stored in a second storage unit of a model storage unit or at least one of a plurality of adapters stored in a third storage unit. Specifically, the computing device may generate the third lens by using a model learning unit to learn at least one of a plurality of foundation models or at least one of a plurality of adapters to be optimized for a data set.

[0182] As a specific example, the computing device may generate a third lens by fine-tuning at least one foundation model based on a data set. Alternatively, the computing device may learn at least one adapter using a data set and generate a third lens comprising the learned adapter and at least one foundation model.

[0183] For example, a computing device can determine a dimension optimized for yielding the intrinsic characteristics of a dataset and train at least one artificial intelligence model to vectorize into a latent space of the determined dimension. As a specific example, the computing device can calculate various metrics for estimating data distribution (e.g., reconstruction error, cluster quality metrics (Silhouette score, Davies-Bouldin Index, etc.), classification or regression accuracy, etc.) and analyze how these metrics change as the embedding dimension changes. For instance, the optimal dimension can be determined by retraining using an autoencoder method while changing the dimension of the latent space to 16, 32, 64, etc., to find the point where the reconstruction error is minimized or the dimension with the highest performance for a specific task (classification, clustering, etc.), but is not limited to this.

[0184] In addition, for example, a computing device can select a candidate model suitable for a specific domain (e.g., medical images, social media text, etc.) from among large-scale pre-trained foundation models (e.g., Vision Transformer, GPT family models, etc.) and perform fine-tuning with a user dataset. Alternatively, by keeping the foundation model fixed and training an adapter (e.g., LoRA, Prompt Tuning, etc.) to update only some of the model parameters, it is possible to reduce computational load while fitting more precisely to the user's data distribution.

[0185] A third lens containing a model trained or tuned in this way generates embeddings specialized for the corresponding dataset, thereby providing higher diagnostic precision and insight.

[0186] The computing device inputs the data set into the third lens and embeds the data set into a latent space of optimal dimensions, thereby forming a second vector set (2 ndA VECTOR SET) can be obtained. Additionally, the computing device calculates at least one characteristic value of the data included in the data set based on the second vector set to obtain a third diagnostic result (3 rd A DIAGNOSIS RESULT can be obtained. That is, the computing device can analyze a second set of vectors (e.g., Euclidean distance, clustering, density calculation, etc.) to produce a third diagnostic result that provides higher-level insights, such as the intrinsic characteristics of the data set, class distribution, or bias.

[0187] A first vector set produced by a second lens (including a pre-trained artificial intelligence model) may be defined based on a predetermined dimension (hereinafter referred to as the "first dimension") according to the structure and parameters of the pre-trained model. On the other hand, a second vector set produced by a third lens (including an artificial intelligence model optimized for the data set) may be defined based on a dimension (hereinafter referred to as the "second dimension") that is adaptively determined by analyzing the intrinsic properties of the data set.

[0188] Accordingly, the first dimension in which the first vector set is defined may differ from the second dimension in which the second vector set is defined. The first dimension in which the first vector set is defined is determined by an artificial intelligence model, and the second dimension in which the second vector set is defined can be adaptively determined according to the dataset. For example, the first dimension according to the second lens reflects the fixed structure of a specific pre-trained model (e.g., 768 dimensions, 1024 dimensions, etc.), but the second dimension according to the third lens can be optimized by considering the manifold structure of the dataset, reconstruction error, or cluster quality indicators.

[0189] Furthermore, according to other embodiments of the present disclosure, since the determination methods of the first and second dimensions are different, the data distribution and characteristics represented by the first vector set and the second vector set may also differ from each other. That is, while embeddings based on a pre-trained model using a second-level lens reflect general and universal features, embeddings customized for a dataset using a third-level lens can more precisely reflect the specificity of the target dataset (bias structure, class boundaries, noise patterns, etc.).

[0190] Accordingly, in a data diagnosis or visualization procedure, the distribution shape, cluster structure, bias indicator, etc. may differ between the result using the first vector set and the result using the second vector set, and the computing device according to the present disclosure may support a balanced identification of the unique characteristics and universal features of the data set by comparing and analyzing the two vector sets.

[0191] In addition, the computing device may provide a vector set corresponding to the data set (e.g., a first vector set or a second vector set) to at least one visualization tool (VISUALIZATION TOOL) to provide a data image (Image of Data, IOD) visualized from the data set to the user.

[0192] At this time, the computing device can generate and output at least one of a plurality of different types of data images (IODs) depending on the type of visualization tool or the dimension visualized by the visualization tool.

[0193] For example, when using a first visualization tool (e.g., PCA, 2D-based dimensionality reduction algorithm), the computing device can project a vector set into two dimensions to obtain a first data image (IOD #1) in two dimensions and then provide it.

[0194] In addition, for example, when using a second visualization tool capable of three-dimensional visualization (e.g., a specific 3D dimension reduction algorithm, a WebGL-based 3D viewer, etc.), the computing device can acquire a three-dimensional second data image (IOD #2) and transmit it to a user device (e.g., a PC, a smartphone, an XR device, etc.) to enable the data distribution to be viewed from a three-dimensional perspective.

[0195] In addition, the third data image (IOD #3) can be obtained by similarly reducing and projecting the dimensions of the data into two, three, or other dimensions using a third visualization tool (e.g., t-SNE, UMAP, etc.).

[0196] In this way, the computing device according to the present invention provides a vector set to various visualization tools to generate multiple forms of data images, thereby enabling the user to intuitively check the characteristics of the data distribution, cluster structure, the location of outliers, and boundaries between classes. In particular, by freely selecting or converting the desired visualization tool and dimension, the user can compare and review different visual representation results (IOD #1, IOD #2, IOD #3, etc.) even for the same data set.

[0197] For example, users can initially observe the overall distribution through 2D visualization and then switch to 3D visualization to more clearly identify spatial adjacency relationships between specific clusters. Alternatively, by applying t-SNE and UMAP alternately to compare the differences between the local structure and the global structure, outliers or biases that pose data quality issues can be identified early.

[0198] As such, "Image of Data (IOD)" in this disclosure refers to a result of visualizing a vector set, but is not limited to specific standards (2D, 3D) or specific techniques (PCA, t-SNE, UMAP, Autoencoder-based visualization, etc.). Furthermore, the computing device may receive user interactions regarding the visualization result in real time (e.g., zooming in and out, clicking on a specific point, specifying a range, changing the color scale, etc.) to update the visualized image of data or additionally display detailed information (e.g., metadata corresponding to each point).

[0199] In addition, the computing device may generate and provide a diagnostic report based on the first diagnostic result, the second diagnostic result, or the third diagnostic result. For example, the computing device may store the first diagnostic result (statistical indicators, ratios of missing values ​​or outliers, etc.), the second diagnostic result (embedding analysis results based on a pre-trained model, cluster distribution maps, distances between classes, etc.), or the third diagnostic result (high-precision embedding and feature analysis results obtained through a custom model or adapter) in a database or manage them in a separate structured format such as JSON or XML, and then combine or arrange them in the form of a dashboard or document to generate a comprehensive report.

[0200] The diagnostic report may include, but is not limited to, information on data quality grades, summary information on factors degrading data quality (e.g., major outliers or causes of bias, indicators of potential label mismatch), recommendations for data improvement (e.g., recommendations for data dieting or bulking up), cluster visualization information (e.g., visualization of major clusters in the embedding space, density or distance measurement results, etc.), or information on learning model performance prediction (e.g., prediction accuracy or F1 Score estimates assuming a specific model).

[0201] These reports can be delivered to users (e.g., data analysts, AI model developers, etc.) in a structured format, and can be immediately displayed as charts, tables, or heatmaps in a graphical user interface (GUI) for immediate use in the analysis or model modification phases.

[0203] FIG. 12 is a diagram illustrating a method for a computing device to construct a data processing model for data imaging and diagnosis according to user input, according to various embodiments.

[0204] Referring to FIG. 12, a computing device or at least one processor included in the computing device may be configured to perform an operation (S1210) of acquiring a first data set. At this time, the computing device may acquire the first data set by receiving the first data set from a user device or by retrieving the stored first data set.

[0205] Additionally, at least one processor may be configured to perform an operation (S1220) of selecting a first level among multiple levels classified according to a data processing method based on user input. Specifically, at least one processor may select a first level according to user input selecting a first level among multiple levels. In this case, at least one processor may determine a first data processing method corresponding to the first level. At least one processor may determine the first data processing method by determining at least one of tools (e.g., an arithmetic unit or an artificial intelligence model, etc.) in which a processing algorithm for processing a first data set is stored.

[0206] Additionally, at least one processor may be configured to perform an operation (S1230) of determining at least one attribute for a first data processing model corresponding to a first level. At this time, the first data processing model may include at least one artificial intelligence model. Specifically, the first data processing model may include at least one pre-trained model stored in advance or an artificial intelligence model trained to be optimized for the first data set. For example, based on an input requesting imaging and diagnosis optimized for the first data set, at least one processor may generate the first data processing model by training an artificial intelligence model optimized for the first data set, but is not limited thereto. Additionally, at least one attribute may include an attribute related to the operation of the first data processing model processing data. For example, at least one attribute may include parameters of the artificial intelligence model, the number of hyperparameters, or graphics card information, but is not limited thereto.

[0207] Additionally, at least one processor may be configured to perform an operation (S1240) of providing prior information related to the task of processing a first data set through a first GUI (Graphic User Interface) using a first data processing model. At this time, the prior information may be information configured to notify the user in relation to the computing device processing the data set to produce results for imaging and diagnosis of the data set.

[0209] FIG. 13 is a drawing for explaining information included in prior information according to various embodiments.

[0210] Referring to FIG. 13, the prior information may include cost information related to the cost incurred in processing the dataset, result preview information predicting and displaying the results of the dataset processing in advance, progress information indicating the progress of the dataset processing, reference result information indicating the processing results of cases similar to the dataset, or attribute information indicating the attributes of the processing model used for data processing. Accordingly, the user can use the service more smoothly by checking the estimated processing cost, predicted results, model progress, etc., in advance at the time of applying for dataset imaging and diagnosis.

[0211] In the present disclosure, "cost information" may refer to information related to the cost incurred in processing a data set. For example, cost information may include an estimate of the computational cost required for processing the data set (e.g., GPU time, CPU core time, memory usage, etc.) or payment costs that a user must pay to use the processing service (e.g., credits, points, currency units, etc.). Specifically, by being presented to the user in a form such as "estimated computation time 2 hours, cost 10 dollars when using 1 GPU," the user can determine in advance how much resources and cost will be consumed depending on the size of the data set, model complexity, etc.

[0212] In the present disclosure, "result preview information" may refer to information that allows a user to gauge the approximate form or value of a result by briefly presenting some key results expected after processing a data set. In this case, the computing device may briefly display the expected accuracy range, cluster distribution preview, or sample visualization images after training an artificial intelligence model. Additionally, for example, the result preview information may include data images obtained by simply inputting the data set into a visualization tool. Although such data images may not accurately reflect the intrinsic characteristics of the data set, they can induce the user to predict results by providing the visualization results of the data set in advance.

[0213] Through this, the user can determine whether the data imaging and diagnostic process of the present disclosure meets their purpose even before obtaining the final result, and, if applicable, can quickly decide to change the direction of analysis or reset options.

[0214] In the present disclosure, "progress information" may refer to information regarding the processing algorithm or processing status of a data set. For example, progress information may include the progress of the optimization learning process based on the data set, the time required for the learning, or the remaining time. By checking the current learning stage (initialization, feature extraction, fine-tuning, validation, etc.) and the estimated remaining time in real time, the user can predict the time when the processing is completed or determine whether to allocate additional resources (e.g., computing resources). Additionally, if a problem (e.g., excessive missing values, a surge in outliers, etc.) is detected during processing, a warning message may be displayed in conjunction with the progress to provide the user with an opportunity to take action in advance.

[0215] In the present disclosure, "reference result information" may refer to information that allows one to indirectly estimate expected results or performance indicators by presenting examples of processing results performed in the past for cases similar to the data set submitted by the user. For example, the reference result information may include result data obtained by imaging and diagnosing reference data similar to the input first data set. For instance, it may be provided in a form such as, "When analyzing 100,000 image data of the same category (domain), the average accuracy was 92%, and the processing time took about 4 hours."

[0216] Through this, users can predict the results of their own data set analysis or formulate processing strategies by referring to insights obtained from similar data processing cases from third parties or others.

[0217] In the present disclosure, "attribute information" may refer to information about attributes associated with a data processing tool. For example, attribute information may include the number of model parameters or graphics card information. For example, it may be indicated as, "This processing uses a Transformer architecture-based model (approximately 100 million parameters) and an RTX 3090 GPU." Through this information, the user can understand the model size, resource compatibility (whether it is an internal GPU or a cloud GPU), development environment, etc., and predict model utilization and training performance in advance.

[0219] Referring again to FIG. 12, at least one processor may be configured to perform an operation (S1250) of constructing a first data processing model based on user input to the first GUI. At this time, the user input to the first GUI may include an approval input for processing a first data set proposed by a computing device.

[0220] That is, the computing device provides prior information to the user regarding the processing of the first data set before processing the first data set, and can process the first data set only when an approval input is received by the user who received the prior information.

[0222] FIG. 14 is a diagram illustrating an example of a computing device constructing a first data processing model according to various embodiments.

[0223] Referring to FIG. 14, a computing device or at least one processor included in the computing device may be configured to perform an operation (S1401) of obtaining a plurality of representative values ​​from a plurality of adapters based on a first data set. At this time, the at least one processor may preprocess the first data set to extract a feature value corresponding to the first data set and input the feature value to each of the plurality of adapters. Additionally, the at least one processor may obtain a plurality of representative values ​​based on data output from each of the plurality of adapters to which the same data has been input.

[0224] Additionally, each adapter that receives a feature value can output a vector set corresponding to the first data set. In this case, at least one processor can obtain a representative value by performing an operation on the output vector set in a predetermined manner. Multiple vector sets output from multiple adapters can be defined based on different dimensions. Multiple adapters can be configured to output vector sets of different dimensions.

[0225] Additionally, at least one processor may be configured to perform an operation (S1402) of selecting a first adapter that satisfies a predetermined condition based on a plurality of representative values. At this time, the predetermined condition may include a condition set to determine an adapter optimized for the first data set. Specifically, at least one processor may select the first adapter by identifying at least one adapter corresponding to the lowest (or highest) value among the plurality of representative values. At this time, the first adapter may be implemented to embed the first data set into an optimal dimension representing the first data set. For example, the first adapter may be configured to output a first vector set of a first dimension based on the first data set.

[0226] Additionally, at least one processor may be configured to perform an operation (S1404) of constructing a first data processing model including a first adapter. Specifically, at least one processor may construct a first data processing model including a foundation model or a pre-training model and a first adapter by communicatingly connecting the first adapter to a pre-stored foundation model or a pre-training model.

[0227] A computing device can build a data processing model optimized for a dataset by additionally training or fine-tuning only the adapter based on the dataset. Since the computing device additionally trains only the adapter on a pre-stored artificial intelligence model, it can consume minimal computational cost.

[0229] FIG. 15 is a drawing illustrating another example in which a computing device constructs a first data processing model according to various embodiments.

[0230] Referring to FIG. 15, a computing device or at least one processor included in the computing device may be configured to perform an operation (S1501) of inputting a first data set to a plurality of foundation models.

[0231] Additionally, at least one processor may be configured to perform an operation (S1502) of obtaining multiple representative values ​​based on data output from multiple foundation models. Specifically, each of the multiple foundation models may output a vector set corresponding to a first data set, and at least one processor may obtain representative values ​​by performing an operation on the output vector set in a predetermined manner. The multiple vector sets output from the multiple foundation models may be defined based on different dimensions. The multiple foundation models may be configured to output vector sets of different dimensions.

[0232] Additionally, at least one processor may be configured to perform an operation (S1503) of determining a first foundation model that satisfies a predetermined condition based on a plurality of representative values, and an operation (S1504) of constructing a first data processing model including the first foundation model. Since the specific method for determining the model based on the predetermined condition and the plurality of representative values ​​has been described above, it will be omitted.

[0234] FIG. 16 is a drawing illustrating another example in which a computing device constructs a first data processing model according to various embodiments.

[0235] Referring to FIG. 16, a computing device or at least one processor included in the computing device may perform an operation (S1601) of inputting a first data set into a first foundation model. At this time, the first foundation model may include a plurality of nodes implemented to receive the same input and output different result values. For example, the first foundation model may include at least one hidden layer comprising a plurality of nodes to which the same input is input in parallel.

[0236] At least one processor can perform an operation (S1602) of obtaining multiple representative values ​​from multiple nodes included in the first foundation model—where the multiple nodes receive feature values ​​for the first data set in parallel. Specifically, the multiple nodes output multiple vector sets based on feature values ​​for the first data set, and at least one processor can obtain multiple representative values ​​from the multiple vector sets based on a predetermined operation.

[0237] Additionally, at least one processor may be configured to perform an operation (S1603) of activating a first node based on a plurality of representative values. Additionally, at least one processor may be configured to perform an operation (S1604) of constructing a first data processing model including a first foundation model in which the first node is activated.

[0238] Specifically, at least one processor can determine a first node that satisfies a predetermined condition based on a plurality of representative values, and can activate the first node so that input is input only to the first node. That is, at least one processor can select a node optimized for the first data set among a plurality of nodes, and can construct a first data processing model by disconnecting communication connections to nodes other than the selected node.

[0240] FIG. 17 is a diagram illustrating a method for imaging and visualizing a data set using a first data processing model in which a computing device is constructed according to various embodiments.

[0241] Referring to FIG. 17, a computing device or at least one processor included in the computing device may input a first data set into the first data processing model after the first data processing model has been built (S1260). Specifically, the computing device may determine that the construction of the first data processing model is complete and generate a trigger signal when the training of an artificial intelligence model for vectorization optimized for the first data set is completed. At least one processor may input the first data set into the first data processing model based on the generated trigger signal.

[0242] Additionally, at least one processor can obtain a first set of vectors defined in a first-dimensional embedding region from a first data processing model (S1270). Each of the plurality of vectors (or data points) included in the first set of vectors may correspond to each unit data included in the first set of data.

[0243] Additionally, at least one processor can obtain multiple characteristic values ​​corresponding to each of the multiple data included in the first data set based on the first vector set (S1280). Specifically, at least one processor can obtain multiple characteristic values ​​corresponding to each of the multiple data based on the distance values ​​between at least two vectors included in the first vector set.

[0244] Additionally, at least one processor may provide a first vector set to a first visualization tool to provide a first data image corresponding to the first data set (S1290). At this time, at least one processor may recommend a visualization tool suitable for visualizing the first data set based on the first vector set or a plurality of feature values.

[0245] For example, according to one embodiment of the present disclosure, a computing device may include a visualization tool DB or communicate with a DB server that manages information related to visualization tools in advance, and perform an algorithm to search for and determine an optimal visualization tool based on data characteristics.

[0246] FIG. 18 is a diagram illustrating a method for a computing device to recommend a visualization tool according to various embodiments.

[0247] Referring to FIG. 18, a computing device or at least one processor included in the computing device may perform a data feature extraction step (S1810). Specifically, the computing device may analyze or extract feature values ​​such as domain, data type, data capacity, vector distribution density, number of items per class, etc. for a data set (or vector set). For example, at least one processor may extract statistical quantities such as data domain, data size (number of samples) and number of dimensions, average distance between vectors, variance, number of clusters, or model training specifications (required computing resources, processing time, etc.) based on the data set.

[0248] These characteristic values ​​can be used internally as arguments when generating SQL queries or as parameters to execute algorithmic logic.

[0249] Additionally, the computing device may perform the steps of querying the visualization tool DB and generating an SQL query (S1820). The computing device (or DB server) may manage metadata for each visualization tool (PCA, T-SNE, UMAP, Autoencoder-based visualization, etc.) in the visualization tool DB (e.g., in the form of a table). At this time, the computing device may generate an SQL query based on matching rules between data characteristics and tool metadata to search for "candidates among available visualization tools that satisfy specific conditions (domain, data size, importance of cluster structure, computation time constraints, etc.)."

[0250] Additionally, the computing device may perform a step of executing a recommendation algorithm and producing a result (S1830). Specifically, when DB search results are returned to multiple visualization tools, the computing device may apply a score calculation or a weight-based algorithm. For example, the computing device may be configured to increase preference for PCA or UMAP when the amount of data is very large, or to increase the score of t-SNE or UMAP when cluster accuracy (fine clustering) is important, or to prioritize PCA, which has low computational complexity, when real-time interaction is required.

[0251] Based on this, the final recommendation priority is determined, and a list such as "t-SNE (1st), UMAP (2nd), PCA (3rd)" can be presented to the user.

[0252] Additionally, the computing device may perform a final decision step (S1840) on a visualization tool based on a user request. For example, the user may be provided with a notice stating that "t-SNE is the most suitable based on data characteristics and priority criteria," and the user may decide whether to use the recommended tool as is or choose another tool.

[0253] According to one embodiment of the present disclosure, even if a user manually specifies a particular tool, they can recognize the major advantages and disadvantages (e.g., visualization quality vs. processing time) in advance through the pre-recommendation results.

[0254] In addition, when a finally determined visualization tool (e.g., a first visualization tool) is applied, at least one processor can generate and provide a first data image by executing the corresponding algorithm on a first vector set. If necessary, it is also possible to compare or analyze multiple data images by sequentially applying other candidate tools (a second, a third visualization tool, etc.).

[0255] In one embodiment of the present disclosure, the computing device incorporates the type of data set (image, text, structured data, etc.), domain (medical, social media, finance, etc.), data volume, distribution characteristics between vectors (density, variance, number of clusters, etc.), processing time constraints, etc., into a "visualization tool recommendation" algorithm, thereby inducing the user to select a visualization tool more rationally. Furthermore, by querying a visualization tool DB with SQL to identify available candidates and calculating the priority among the candidates, the user can obtain visualization results that enable efficient understanding of the high-dimensional embedding space.

[0256] In this way, the computing device according to the embodiment of the present disclosure provides an environment in which data distribution or cluster structure can be accurately and intuitively understood by appropriately utilizing the advantages and disadvantages of each of the techniques such as PCA, T-SNE, and UMAP through a visualization tool recommendation process based on data characteristics.

[0258] FIG. 19 is a diagram illustrating a method for a computing device to generate synthetic data based on data imaging according to various embodiments.

[0259] Referring to FIG. 19, a computing device or at least one processor included in the computing device can generate synthetic data based on a vector set corresponding to a dataset.

[0260] Specifically, at least one processor can acquire a vector set by imaging a dataset using a lens including at least one artificial intelligence model, and can display a data image (IOD) corresponding to the dataset in a visualization space by visualizing the dataset using at least one visualization tool.

[0261] At this time, at least one processor can generate synthetic data based on a set of vectors. Specifically, at least one processor can input at least one vector included in the set of vectors into a calculator with predetermined operation conditions set. In this case, the calculator can output a latent code based on at least one vector. Additionally, at least one processor can input the output latent code into a generative model. At this time, the latent code may be an arbitrary vector defined in a latent space (or embedding space). The generative model may include at least one artificial intelligence model for data generation (e.g., Diffusion model, GAN, VAE, etc.). The generative model can generate and output synthetic data based on the input latent code.

[0262] At least one processor can visually represent the relationship between the synthetic data and the existing data set by visualizing the generated synthetic data. Specifically, at least one processor can input the synthetic data into a visualization tool to display synthetic points (SP) corresponding to the synthetic data on the data image. Alternatively, at least one processor can input the synthetic data into a lens to obtain a synthetic vector (SYNTHETIC VECTOR, not shown) corresponding to the synthetic data, and input the synthetic vector into a visualization tool to display synthetic points corresponding to the synthetic data on the data image.

[0263] By visually representing the location of synthetic data generated from data images of an existing dataset, users can be guided to intuitively recognize the relationship between the generated data and the existing data.

[0265] FIG. 20 is a diagram illustrating a method for a computing device to generate synthetic data based on user input according to various embodiments.

[0266] FIG. 21 is a diagram illustrating an example of a method in which a computing device generates synthetic data based on user input according to various embodiments.

[0267] Referring to FIG. 20, a computing device or at least one processor included in the computing device can visualize a data set and provide a first data image corresponding to the data set (S2010). For example, by utilizing visualization tools such as the aforementioned PCA, t-SNE, and UMAP, each unit data included in the data set can be projected into a two-dimensional or three-dimensional space, and then a data image (IOD) including a plurality of data points can be displayed on a GUI (Graphic User Interface).

[0268] Additionally, at least one processor can receive user input for a first area on the first data image (S2020). For example, referring to FIG. 21, input such as a click, touch, or drag by a user can be recognized for a first area (R) representing a specific point or range on the first data image (IOD). Here, user input is an input event that occurs on the GUI and encompasses various forms such as a left mouse click, mobile touch, pen drawing, etc.

[0269] Referring again to FIG. 20, at least one processor can determine a latent code corresponding to a first region (S2030). The latent code is used as an internal model embedding or noise vector when generating synthetic data.

[0270] At this time, at least one processor may request feedback regarding a first area specified by user input. Specifically, at least one processor may provide the user with a GUI (e.g., a pop-up window or highlighting) that visualizes the first area or displays a confirmation message (e.g., "Is this area the target area for generating synthetic data?"). This allows for error prevention by enabling a cancellation or reset procedure in the event that a point not matching the user's intent is clicked incorrectly.

[0271] At this time, at least one processor can define conditions for data generation based on user input. Specifically, at least one processor can define data generation conditions for determining a latent code to be input to a data generation model.

[0272] Additionally, at least one processor can determine a latent code in various ways depending on the attributes of the first region. For example, at least one processor may selectively perform an algorithm for determining a latent code based on the presence or absence of data points included in the first region or the number of data points.

[0273] For example, if the first region does not contain a data point, at least one processor may determine a potential code based on at least one vector corresponding to at least one point adjacent to the first region. Alternatively, at least one processor may derive a target region including the first region where user input is made and at least one point adjacent thereto, and then determine a potential code based on at least one vector corresponding to at least one point included in the target region.

[0274] As another example, if at least one data point is included in the first region, at least one processor can determine a potential code based on at least one vector corresponding to at least one data point included in the first region.

[0275] At this time, at least one processor can determine a latent code based on at least two vectors. Specifically, at least one processor can identify at least two vectors corresponding to at least one point adjacent to a first region and determine a latent code corresponding to the first region based on the identified at least two vectors. For example, if a vector interpolated by vectors corresponding to points near the first region is set as the latent code, synthetic data reflecting centroid-like characteristics of the region can be generated.

[0276] In addition, if a single point is included in the first region, it can be defined as a target vector, and a latent code can be generated based on the relationships (distance, density, label, etc.) between adjacent vectors around this target vector.

[0277] As another example, if a cluster of multiple data points exists in the first area, at least one processor can determine a potential code based on the attributes of the cluster. Specifically, referring to FIG. 21, if a user drags and selects a portion of a specific dense cluster on the IOD, a potential code can be determined based on centroid or boundary information of the data points belonging to that cluster.

[0278] Referring again to FIG. 20, at least one processor can generate synthetic data by providing a determined latent code to a generation model (S2040). Additionally, at least one processor can input the generated synthetic data into a data processing model to obtain a synthetic vector corresponding to the synthetic data (S2050). Additionally, at least one processor can visualize the synthetic vector and provide a synthetic point corresponding to the synthetic data on a first data image (S2060).

[0279] At least one processor can identify at least one area requiring improvement based on diagnostic results for a data set. At least one processor can visually display the identified at least one area through a display on a user device.

[0280] At least one processor can activate at least one identified area to an interactive state. Specifically, at least one processor can set at least one area requiring improvement to a state where user input is possible. In this case, based on user input regarding at least one area requiring improvement, a data improvement process for the area requiring improvement can be performed.

[0281] At least one processor can extract at least one feature of multiple unit data included in a dataset based on a set of vectors in which the dataset is embedded in a latent space of a specific dimension. At least one processor can identify data that needs improvement based on the extracted at least one feature and can visually represent a data point corresponding to the data that needs improvement or a specific area containing the data point.

[0283] Referring to FIG. 21, at least one processor can generate synthetic data by further incorporating user prompt input. Specifically, at least one processor can receive user prompts through an input interface provided on a GUI (e.g., text field, voice input, slider, etc.). User prompts can express topics, styles, attributes, constraints, etc., regarding the synthetic data in natural language or based on tags. For example, user prompts may include text instructions in the form of “Generate a cat image on a blue background” or “Silhouette format with emphasized perspective.” Additionally, if user prompts specify numeric parameters or categorical tags (e.g., “realistic”, “cartoon”), the corresponding values ​​may be interpreted as internal prompt information and passed to the model.

[0284] At least one processor can combine a determined latent code and a prompt input by the user to set as a condition for a generative model (e.g., GAN, VAE, Diffusion Model, etc.). For instance, the latent code is used as a base latent vector for image generation, and the prompt can be converted into a text embedding (e.g., CLIP, BERT-based encoder) or a conditional layer and input into the model. This structure can be implemented similarly to how the entire synthesis process is conditioned through text embeddings in a "text-to-image" type of diffusion model (e.g., Stable Diffusion).

[0285] In this case, the generative model determines the basic shape or feature distribution based on the latent code, additionally reflects themes, styles, or detailed attributes based on prompts (text, tags, parameters, etc.), and can output the final synthetic data. For example, if the latent code is the result of "interpolation between two vectors," it already reflects intermediate attributes (e.g., mixing two classes) or spatial characteristics; and if the prompt is given as "cat + blue background + cartoon style," the style or thematic elements are reflected through an internal conditioning path within the model, so that a synthetic image with those characteristics can be output.

[0286] A prompt entered into the input interface via the GUI, along with a determined latent code, can be set as a constraint for data generation. At least one processor can input the latent code and the prompt into a generation model, and the generation model can generate synthetic data conditioned by the matters directed by the prompt based on the latent code.

[0287] In addition, at least one processor can convert the user prompt into a vector form based on a text embedding module (CLIP, BERT, GPT, etc.) or a conditioning network (condition layer, cross-attention, etc.) and then cross-attention the hidden layer of the generative model or inject it as a conditional input.

[0288] In addition, at least one processor can be implemented to reflect user requests at each step (e.g., diffusion step, GAN upsampling step, etc.) based on logic for combining latent codes and text embeddings (e.g., concat, add, attention).

[0289] According to one embodiment illustrated in FIG. 21, after checking the synthesized result, the user can perform iterative (refinement) synthesis by modifying the prompt or resetting the potential code (interpolation parameters, additional vector selection, etc.).

[0290] According to an embodiment of the present disclosure, customized synthetic data can be easily obtained by immediately reflecting the conditions desired by the user, compared to when only a latent code obtained by simply interpolating multiple vectors is used.

[0291] In addition, according to an embodiment of the present disclosure, since the latent code already reflects the inherent characteristics of the data set, prompt-based generation is advantageous for generating distribution-preserving results.

[0292] In addition, according to an embodiment of the present disclosure, by efficiently testing various scenarios (adversarial examples, artistic expressions, etc.) through prompt changes, creative utilization of the artificial intelligence model (e.g., content creation, data augmentation, simulation, etc.) becomes easier.

[0294] In addition, at least one processor can automatically extract or generate a prompt from a first area specified by the user, thereby utilizing the said prompt as a condition for generating synthetic data. Through this, a synthetic result reflecting the intrinsic attributes or meta-information of the data can be obtained without the user having to directly input separate text.

[0295] For example, at least one processor can automatically generate a prompt by analyzing label information, class / category, metadata (e.g., time, location, ID, etc.), visual features (in the case of an image), etc., for data points belonging to or adjacent to a first region. Specifically, if it is inferred that it is an "area where cat faces are clustered," at least one processor can generate a simple phrase such as "cat face cluster" or more specific text such as "Multiple cat faces in a close-up shot."

[0296] At least one processor can generate prompts using image captioning technology (e.g., vision-language models, OCR, etc.), and for structured data, can automatically convert category names or key attribute names into sentence form.

[0297] As a specific example, at least one processor can use an image captioning model (e.g., CNN+LSTM structure, Transformer-based visual-language model, etc.) to take an image of the first region as input and automatically generate a sentence-form description (caption). A GUI interface may be provided to the user to "view and edit the extracted text," and the user may manually modify the text as desired and adopt it as the final prompt.

[0298] At least one processor can input a determined latent code or other vector interpolation result and an automatically generated prompt into a generative model (GAN, VAE, Diffusion, etc.).

[0299] In this case, the generative model can output conditional synthetic data by reflecting the data distribution inferred from the latent code, while also reflecting context, object information, attributes, etc. extracted from the user domain.

[0300] As such, the computing device according to the embodiment of the present disclosure generates linguistic constraints along with latent codes, thereby enabling "dimensional expansion" (generation of data with attributes that cannot be obtained with existing data). For example, the computing device can construct an N+3 dimensional data lens by additionally learning synthetic data generated based on linguistic constraints for a data lens that outputs a vector in an N-dimensional latent space.

[0302] According to an embodiment of the present disclosure, through an automatic prompt generation function, user input is simplified and an intuitive synthetic data generation flow is ensured.

[0303] In addition, according to an embodiment of the present disclosure, by converting meta-information, visual information, labels, etc. of a user-specified area into "text conditions," the inherent meaning of the data is naturally reflected.

[0304] Furthermore, according to an embodiment of the present disclosure, by linking additional models such as captioning or OCR to perform advanced object recognition or context description, it is possible to flexibly apply it to various domains (e.g., medical imaging, satellite imagery, document processing, etc.).

[0305] Accordingly, according to an embodiment of the present disclosure, a prompt is automatically extracted or generated from a user-specified area and input into a generation model along with a previously determined latent code, thereby enabling the acquisition of conditional synthetic data that reflects the attributes of the area.

[0307] A computing device according to one embodiment of the present disclosure can perform an interaction operation to remove at least some of the data included in a data set based on user input.

[0308] FIG. 22 is a diagram illustrating a method for a computing device to remove data based on user input according to various embodiments.

[0309] Referring to FIG. 22, a computing device or at least one processor included in the computing device may receive user input regarding a first region comprising at least one point on a visualized data image (S2210). For example, the first region may be a dense region (a part with very high data density within a cluster) or an region that the user wishes to remove, such as a region containing many outliers. At least one processor may provide the user with information regarding a region where data removal is required (e.g., a dense region or an outlier region), and the user may provide input regarding the region. The first region may include at least one cluster in which the characteristics of the data satisfy predetermined conditions.

[0310] In addition, at least one processor can remove at least some of the data corresponding to multiple points included in the first region (S2220). Here, "at least some" refers to only the data corresponding to the entire or specific conditions (e.g., outlier filter, specific label, etc.) specified by the user.

[0311] At least one processor may selectively perform a data removal algorithm based on the attributes of a first region. For example, if at least one processor determines that data clusters included in the first region are excessively dense and cause imbalance, it may undersample all data points within that region or remove them at an arbitrary rate. Additionally, for example, at least one processor may selectively remove only data points within that region that satisfy predefined outlier detection criteria (distance, density, label mismatch, etc.) by deciding to remove only outliers within the first region. In this case, at least one processor may provide refinement options (filter conditions, etc.) to remove only points that fall outside a specific label or a specific statistical range, even within a user-defined region. (e.g., "Remove only points that belong to this region and have a label of 0, etc.")

[0312] Additionally, at least one processor can provide a data image from which at least some data has been removed (S2230). While viewing the newly visualized distribution, the user can intuitively understand the effect of improving data imbalance or reducing noise. If necessary, a deletion history may be recorded or an undo function may be provided to prevent data loss due to user error.

[0313] In addition, at least one processor can perform reimaging and diagnostics (cluster analysis) on the removed dataset to verify how much the data quality has improved and whether there is an advantage in model training.

[0314] According to an embodiment of the present disclosure, since direct removal is possible through user interaction, detailed domain knowledge that existing automated algorithms may miss can be reflected.

[0315] In addition, according to an embodiment of the present disclosure, substantial advantages are provided, such as resolving data imbalance or preventing overfitting during training, through undersampling of dense regions or removal of outliers.

[0316] In addition, according to an embodiment of the present disclosure, immediate feedback can be obtained in a visualized (2D or 3D) space, allowing for quick determination of subsequent actions such as "how much data has disappeared" or "how the distribution has changed."

[0317] Accordingly, according to an embodiment of the present disclosure, by enabling the removal of data in a user-specified area (particularly cluster-dense areas, outlier-frequent areas, etc.) in an interactive manner, an intuitive and flexible tool for data quality management can be provided.

[0319] FIG. 23 is a diagram illustrating a method in which a computing device, according to various embodiments, performs data improvement in response to a data improvement request and provides a visual interaction therewith.

[0320] Referring to FIG. 23, a computing device or at least one processor included in the computing device may receive a data improvement request (S2310). Specifically, at least one processor may recognize a data improvement request by receiving user input through at least one GUI that directs data improvement. For example, if a button to improve data in a first manner (e.g., "resolve data imbalance") or a button to improve data in a second manner (e.g., "remove noise area") is selected in the user GUI, at least one processor may recognize that an improvement request has occurred through such input. Additionally, the user may recognize that a specific class (or attribute) is lacking and request, such as "create more of that class."

[0321] Additionally, at least one processor can visually represent at least one area requiring data improvement (S2320). Specifically, at least one processor can acquire characteristics of a data set based on a vector set corresponding to the data set, and detect an area requiring data improvement based on the characteristics of the data set.

[0322] For example, at least one processor can identify areas where a specific class is too scarce or where certain areas (clusters) are excessively dense by analyzing the density of the data set. Additionally, for example, at least one processor can automatically identify points where a specific class is insufficient or where a specific class is over-dense by analyzing the bias of the data set, such as when the class distribution is unbalanced and needs improvement. Additionally, for example, at least one processor can detect areas where a large number of outliers exist (areas where noisy data is concentrated) by performing outlier analysis.

[0323] At least one processor can notify the user of the detected "areas requiring improvement" by highlighting them on the visualized data image (IOD) or emphasizing them with borders. In this case, at least one processor can provide an interactive guide (e.g., tooltip) on the GUI regarding the criteria used to select the areas. By checking this indication, the user can intuitively identify the reasons why data improvement is needed (e.g., "where data is insufficient, excessive, or noise has occurred").

[0324] Additionally, at least one processor can perform data improvement by generating or removing data for each of at least one region (S2330). For example, at least one processor can generate synthetic data based on the data generation method according to FIG. 19 for a first region on a data image where it is determined that data of a specific class is insufficient. Additionally, for example, at least one processor can remove at least some data based on the data removal method according to FIG. 22 for a second region where data is overcrowded or contains outliers. At this time, the user can select at least one of various detailed options (e.g., execute, cancel, adjust detailed options, etc.) for the automatic suggestion.

[0325] Additionally, at least one processor may visualize and provide an improved data image corresponding to the result of the data improvement (S2340). This allows the user to see at a glance how the data distribution has changed and intuitively understand the extent to which imbalances have been resolved and how much outliers have been reduced. Optionally, at least one processor may also provide the user with a comparison view with the immediate previous state (before / after visualization) or a diagnostic report.

[0326] Additionally, the user can perform manual corrections by deselecting some of the visualized improvement areas or by instructing the expansion of the scope of creation or removal. Furthermore, at least one processor can automatically identify necessary areas based on the improvement results and perform a loop to repeatedly improve data quality multiple times.

[0328] FIG. 24 is a drawing illustrating an example of a computing device in which a data processing method including a snapshot function is implemented, according to various embodiments.

[0329] Referring to FIG. 24, a computing device (2400) may include a plurality of configurations for obtaining snapshot information based on a specific scene on a data image. Here, the plurality of configurations are arbitrarily separated to perform a specific operation, and may be physically separated configurations, or may be configurations separated based on various operations implemented in a single software program.

[0330] Specifically, the computing device (2400) may include a screener for screening a data image or a vector set corresponding to the data image, an event detector for detecting whether an event to take a snapshot has occurred, a capture unit for taking a snapshot, a modifier for correcting the captured scene, and a generator for generating various information about the snapshot.

[0331] The screener can screen a vector set or a data image. Specifically, the screener may have at least one metric set for measuring characteristic values ​​of the data set (e.g., density, bias, homogeneity, presence of outliers, etc.) based on the vector set or the data image. For example, the screener may analyze basic characteristics (density, outlier ratio, etc.) in a 2D / 3D data image based on a first metric, or measure high-dimensional distribution characteristics in a high-dimensional vector set based on a second metric.

[0332] An event detector can detect events occurring during screening. Specifically, the event detector can monitor measurement results from the screener and detect that an event has occurred if the monitored results satisfy predetermined conditions. In this case, the event detector can store information about the data where the event occurred (e.g., data item, vector value corresponding to the data, coordinates of the data point corresponding to the data, etc.). For example, if the event detector detects a specific area where the data density is greater than a threshold or where the bias indicator is above a reference value based on the screener results, the event detector recognizes this as an "event occurrence" and can generate a signal.

[0333] The generator can generate information regarding events detected by the event detection unit. Specifically, the generator can generate tag information by synthesizing event information detected by the event detector, the location and metadata of the captured scene, and characteristic values ​​calculated by the screener. Alternatively, the generator can record additional information related to the snapshot (class label, time, user ID, etc.). That is, the generator can generate tag information based on information about data where an event occurred that is either stored in advance or generated by the event detector. The tag information generated by the generator may include, for example, identification labels such as "areas with unusual characteristics (outlier concentration)" or "sections with excessively high density," the time of snapshot capture, and key characteristic values.

[0334] Thus, the computing device can enable the user to immediately grasp the context when retrieving the snapshot later.

[0335] The capture unit can capture and save a specific scene on the data image. For example, when a signal indicating the occurrence of an event is transmitted from an event detector, or when a user commands the capture unit to take a snapshot directly, the capture unit can capture (take) a scene containing at least one data point corresponding to the event on the data image at a specific viewpoint. At this time, "capturing" includes the process of saving the state of an actual GUI screen (2D / 3D view) or an internal data structure (vector set + visualization mapping parameters) as an image (or video frame).

[0336] The correction unit can perform modifications on the captured scene (snapshot). This allows for providing the user with a highly visible result, and if necessary, can also perform graphic corrections such as visually highlighting areas or adding labels.

[0337] For example, a computing device may provide a vector set corresponding to a data set to a screener, and the screener may diagnose the characteristics of the data set based on the vector set. An event detector may identify whether an event has occurred based on the characteristics diagnosed by the screener. For example, if the characteristics diagnosed by the screener satisfy a predetermined condition, the event detector may generate a signal indicating that an event has occurred. In this case, the capture unit may capture a snapshot corresponding to the event based on the signal indicating that the event has occurred. Specifically, the capture unit may generate a snapshot by capturing a scene containing at least one data point corresponding to at least one vector associated with the event at a specific point in time. In addition, in parallel with this, a generator may generate tag information based on information about the event that occurred or information about the location on the data image where the event occurred. Additionally, a correction unit may correct the generated snapshot in a predetermined manner (e.g., brightness adjustment, contrast adjustment, etc.).

[0339] FIG. 25 is a diagram illustrating a method for a computing device to screen a data set and generate snapshot information according to various embodiments.

[0340] Referring to FIG. 25, a computing device or at least one processor included in the computing device can screen a vector set by using a screener in which at least one metric is set, and by calculating at least one characteristic value based on at least one vector included in the vector set (S2510). Specifically, at least one processor can diagnose the characteristics of the data and measure various indicators such as the density, bias, and homogeneity of the data.

[0341] In the present disclosure, screening may proceed along a specific path based on a specific starting point (coordinate location) on a data image (2D / 3D) or a vector set. For example, at least one processor may evaluate the density, bias, homogeneity, etc. of the data distribution by sequentially scanning and analyzing a certain range from a predetermined location or a determined reference point.

[0342] Additionally, at least one processor may perform screening of multiple levels distinguished according to the screening target. Specifically, at least one processor may perform at least one of a first level screening for screening data images or a second level screening for screening vector sets. The difference is that the first level screening performs operations on two-dimensional or three-dimensional data based on two-dimensional or three-dimensional data images, whereas the second level screening performs high-dimensional operations based on high-dimensional vector sets.

[0343] For example, when a first level of screening is indicated, at least one processor may apply a first metric to measure the characteristics of the data by performing an operation (two-dimensional or three-dimensional data operation) based on at least one data point included in the data image.

[0344] Additionally, for example, when a second level of screening is directed, at least one processor may apply a second metric to measure the characteristics of the data by performing an operation (high-dimensional operation) based on at least one vector included in the vector set. For example, at least one processor may diagnose what structure (clusters, outliers, etc.) the vectors to be screened form on an actual high-dimensional manifold, how high the bias indicator is, etc.

[0345] In addition, not limited thereto, at least one processor may perform screening of multiple levels classified according to the type of characteristic to be screened and diagnosed. Specifically, at least one processor may perform at least one of a first level screening in which a metric is set for measuring a first type of characteristic (e.g., missing values, data statistics, etc.), a second level screening in which a metric is set for measuring a second type of characteristic (e.g., density, etc.), or a third level screening in which a metric is set for measuring a third type of characteristic (e.g., class distribution, etc.).

[0346] Additionally, at least one processor can identify at least one vector in which at least one characteristic value satisfies a predetermined condition (S2520). Through this, at least one processor can identify at least one vector in which a first characteristic (e.g., density) satisfies a first condition (e.g., overcrowded area, etc.) or at least one vector in which a second characteristic (e.g., bias index) satisfies a second condition (e.g., bias degree is above a standard).

[0347] Additionally, at least one processor can obtain a data image including a plurality of data points representing the data set in two or three dimensions by processing the vector set using at least one visualization tool (S2530).

[0348] Additionally, at least one processor can determine a target region including at least one data point corresponding to at least one vector on a data image and generate tag information associated with the target region (S2540).

[0349] Specifically, at least one processor can identify at least one data point corresponding to at least one vector satisfying a predetermined condition, and can determine a target area by determining a predetermined area with at least one data point as a reference point. For example, at least one processor can determine an area having a predetermined radius centered on at least one data point as the target area.

[0350] Additionally, at least one processor may generate tag information associated with a determined target area. In this case, the tag information may reflect the diagnostic results associated with the target area. For example, at least one processor may generate tag information based on information about an event detected by an event detector and characteristic values ​​diagnosed by a screener, but is not limited thereto. For example, the tag information may include the background of discovery (e.g., which metric conditions were met), time, analyst ID, event type (e.g., anomalous interval, blank interval, overcrowded interval, etc.), but is not limited thereto.

[0351] Additionally, at least one processor may store snapshot information including tag information and a first scene containing a target area—the first scene is generated by capturing the target area from a specific viewpoint (S2550). Specifically, at least one processor may acquire the first scene by capturing a view in which the target area is most clearly visible visually. At least one processor may acquire the first scene by selecting the optimal scene among a plurality of captured scenes while adjusting the viewpoint of a virtual camera for capturing the target area. Alternatively, a computing device may automatically calculate the camera angle / magnification that best reveals the target area, adjust the viewpoint, and then take a snapshot (e.g., 3D point cloud rotation).

[0352] In some cases, at least one processor may compare multiple candidate scenes (e.g., top view, side view, or 45-degree angle view) and provide them to the user, and determine the scene selected by the user as the first scene. Additionally, the captured snapshot (first scene) may be corrected (brightness, brightness, highlighting, etc.) by the correction unit (MODIFIER) described above, and the snapshot information including the corrected final scene and tag information may be stored in a DB or file format.

[0354] Referring again to FIG. 24, the computing device (2400) can generate snapshot information based on user input received through a GUI-based input interface. Specifically, the computing device (2400) transmits a signal instructing the capture unit to generate a snapshot based on user input, and the capture unit can capture at least a portion of the data image (IOD).

[0355] FIG. 26 is a diagram illustrating a method for a computing device to provide snapshot information based on user input according to various embodiments.

[0356] Figure 27 is an example of a screen provided by a computing device.

[0357] Referring to FIG. 26, a computing device or at least one processor included in the computing device can provide a first data image corresponding to a first data set through a first view port (S2610).

[0358] For example, referring to FIG. 27, the computing device may provide a first data image corresponding to a first data set through a first view port (2710). At this time, the computing device may also provide preview information for at least some of the plurality of data points included in the first data image.

[0359] For example, a computing device can provide preview information by outputting actual data corresponding to a data point. Specifically, the computing device can configure the preview information so that clicking on a specific data point allows the user to preview the actual data corresponding to that point (e.g., original image, text content, statistical value, etc.). Through this, the user can immediately verify the meaning represented by the point on the data image.

[0360] Referring again to FIG. 26, at least one processor can generate first snapshot information including a first scene of a first data image being provided through a first view port in response to user input received through a first GUI and provide it through a second view port (S2620). Specifically, at least one processor can generate the first snapshot information by capturing a first scene including at least one data point on the first data image.

[0361] For example, referring to FIG. 27, the computing device may receive user input for the first GUI (2720) and provide first snapshot information including a first scene for the first data image through the second viewport (2730). Specifically, the computing device may display the first GUI (2720) and induce the user to specify a desired scene using a button or a specific gesture (such as dragging or box selection) that instructs the user to take a snapshot.

[0362] Additionally, if a user requests a "snapshot," the computing device may transmit an instruction signal to the capture unit. In this case, the capture unit may capture a first scene by combining the current viewpoint, magnification ratio, or user range specification of the first view port (2710) in which the first data image is displayed. Subsequently, the computing device may provide the completed first snapshot information to the user through the second view port (2730). In the example of FIG. 27, the second view port (2730) may be used as a "snapshot preview" area.

[0363] Additionally, at least one processor may further provide an interface for receiving memo input from a user through a second view port. For example, the computing device may enable a "memo input window" (not shown) within the second view port (2730). While viewing the snapshot, the user may directly write string comments (memos) (e.g., "Class A is excessively dense here," "suspected outlier," etc.). Additionally, at least one processor may save the user memos by mapping them to the snapshot information, so that when the snapshot is re-examined later, the memo information written by the user can be retrieved together.

[0364] Additionally, at least one processor may further provide an interface for generating tag information for the first snapshot information through the second viewport. Specifically, when the first snapshot information is generated by a user, at least one processor may activate an interface requesting the selection of at least one of a plurality of tags through at least a part of the second viewport. In this case, at least one processor may determine a recommended tag based on the captured scene and provide information about the recommended tag to the user. Additionally, at least one processor may generate and store tag information based on the selected tag. At least one processor may structure and store the snapshot information according to the tag type.

[0365] For example, the computing device may display a "tag selection menu" (not shown) in a portion of the second view port (2730) (e.g., popup, side panel). At least one processor may analyze the captured scene (such as data distribution or point attributes within the snapshot) and automatically suggest recommended tags such as "bias," "clustering," "outliers," and "label mismatch." When a user selects one or more tags, the processor may generate and store final tag information (e.g., "clustering," "bottom right cluster," "Class A") based on this.

[0367] Referring again to FIG. 26, at least one processor can generate link information for connecting the first snapshot information to an external communication network based on user input for the second GUI provided through the second view port (S2630).

[0368] For example, referring to FIG. 27, at least one processor may provide a second GUI (2740) through a second viewport (2720). At least one processor may generate link information based on user input to the second GUI (2740). The link information may include a link for transmitting the first snapshot information through an external communication network. The link information may serve as a connection path for transmitting and sharing this snapshot through an external communication network (Internet, company internal network, etc.), and the recipient may view the same scene (snapshot) and tag information or memo information, etc., through the link.

[0369] According to the embodiments of the present disclosure, since the user directly determines and captures an arbitrary point in time / area, customized scenarios that may be missed in automatic capture (such as greatly enlarging only specific clusters or highlighting only specific labels) are effectively supported.

[0370] In addition, according to an embodiment of the present disclosure, user comments or classification information are organically combined with snapshots and stored through viewport switching (from the first to the second) and a memo / tag interface, thereby providing a richer context for future analysis collaboration or document reporting.

[0371] In addition, according to the embodiments of the present disclosure, the link generation function enables real-time sharing and feedback of snapshot information (scenes, tags, memos, etc.) with external team members or other systems, improves the efficiency of data interpretation in a remote collaboration environment, and can have economic ripple effects by inducing external exposure of the solution.

[0373] FIG. 28 is a diagram illustrating the function of a computing device to replay snapshot information according to various embodiments.

[0374] A computing device (or at least one processor) according to an embodiment of the present disclosure can sequentially store a plurality of snapshots captured along a screening path and continuously provide these snapshots according to user playback instructions, thereby enabling playback in the form of an image or a slide.

[0375] Referring to FIG. 28, a computing device or at least one processor included in the computing device may sequentially store multiple snapshot information obtained along a screening path (S2810). Specifically, a screener included in the computing device may search a data image (or a high-dimensional vector set) along a specific path (referred to as a "screening path"), and may take snapshots at each point when a specific characteristic is detected, an event occurs, or at regular intervals. For example, screening may proceed by "inspecting each block (region) on a 2D view while gradually moving from the top left to the bottom right." After checking the characteristic value at each inspection point, a snapshot may be taken if it exceeds a threshold value.

[0376] Additionally, at least one processor may record the snapshots captured in this manner along with connection information (e.g., "Snap #1 to Snap #2 to Snap #3 ...") according to the order of acquisition or the order along the screening path. In this case, each snapshot may be stored including metadata such as "view point, location of capture (coordinates along the screening path or high-dimensional mapping information), cause of event (over-density, outlier detection, etc.), and time of capture."

[0377] Thus, snapshots can be listed and saved in the order in which the scanning proceeded, and later, the user can visually review the continuous exploration process by playing back the sequence of those points in time.

[0378] Additionally, at least one processor can play back multiple snapshot information stored sequentially in a predetermined manner according to a playback instruction entered by a user (S2820). Specifically, when a user (such as an analyst) presses "Play" or inputs a request instructing to sequentially check screening records through a specific interface, at least one processor can retrieve multiple snapshot information stored sequentially.

[0379] At this time, the computing device may be implemented such that the transmission order during playback of multiple snapshot information differs from the actual snapshot capture order. For example, the user may take snapshots in a random order at the time of shooting, but may rearrange them according to specific criteria (e.g., problem priority, reverse chronological order, user-defined order, etc.) during playback.

[0380] Additionally, at least one processor may output snapshots by applying a predetermined playback method, such as a slideshow (continuous screen transitions at fixed intervals), animation (smooth transitions between scenes), or timeline motion (proceeding in sync with the occurrence of events in each snapshot). At this time, at least one processor may visually display at least one area where a snapshot was captured on the data image and display a movement path connecting those points in the form of an animation on the screen. For example, the computing device may provide a dynamic exploration experience to the user by creating a visual effect that appears as if a camera is moving through the areas where snapshots were taken in sequence.

[0381] At this time, the computing device can receive user commands related to playback and control playback according to the received commands. For example, at least one processor can perform actions such as "pause," "skip," "reverse," and "adding comments to individual snapshots" based on interaction actions by the user.

[0382] Through this, specific snapshots during playback can be examined in more detail, and data enhancement actions, such as removing or synthesizing data at that point if necessary, can be determined. Specifically, at least one processor may induce the user to use the enhancement function by providing enhancement information (e.g., the reason why enhancement is needed) when an area requiring enhancement is identified during the playback of a snapshot.

[0384] A computing device according to an embodiment of the present disclosure may provide a function for linking snapshot information and a diagnostic report. Specifically, the computing device or at least one processor included in the computing device may generate a diagnostic report describing the characteristics of a data set along with snapshot information for a specific area during a data screening process.

[0385] FIG. 29 is a diagram illustrating a method in which a computing device generates diagnostic reports and snapshot information in conjunction according to various embodiments.

[0386] Referring to FIG. 29, a computing device or at least one processor included in the computing device can perform a data screening operation (S2910). Specifically, at least one processor can use a screener to measure specific metrics (density, bias, outlier ratio, etc.) on a data set (or vector set) and produce a diagnostic result (e.g., overcrowding in a specific area, class imbalance).

[0387] Additionally, at least one processor can perform a diagnostic report generation operation (S2920). Specifically, at least one processor can generate a diagnostic report by aggregating or summarizing various diagnostic results. Since the method for generating the diagnostic report has been described above, a detailed explanation will be omitted.

[0388] Additionally, at least one processor can perform a snapshot information generation operation (S2930). Specifically, if an unusual area (overcrowded area, concentration of outliers, etc.) is detected during the screening process, at least one processor can capture the scene and generate a snapshot.

[0389] At this time, if detailed diagnostic results are already prepared in the form of a report, at least one processor may record snapshot information in a shortened or summarized form of the diagnostic results (e.g., "20 noise points found, cluster with bias index of 0.85 or higher"). As a specific example, at least one processor may generate snapshot information based on summary information for a specific diagnostic result and a scene capturing an area on a data image corresponding to that diagnostic result.

[0390] Alternatively, when a diagnostic report is generated during the screening process, at least one processor may convert parts of the diagnostic results that satisfy specific criteria (e.g., above a threshold, class of interest, etc.) into snapshot information.

[0391] At least one processor can perform a linkage operation (S2940) between the diagnostic report and the snapshot information. Specifically, at least one processor can link and store the diagnostic results on the diagnostic report with the snapshot information corresponding to the diagnostic results. As a specific example, when at least one processor stores the two data (snapshot information, diagnostic report) simultaneously in a database or file system, it can link them by mutually linking unique IDs (e.g., report ID, snapshot ID) or by assigning the same metadata (e.g., time, area coordinates, event type).

[0392] At least one processor can perform a diagnostic report retrieval operation (S2950) from a snapshot. Specifically, in a GUI where snapshot information is displayed, if a user clicks a button instructing to retrieve a diagnostic report or clicks a label (such as class imbalance) displayed on the snapshot, the computing device can quickly retrieve a diagnostic result corresponding to the snapshot based on linked stored information. Subsequently, at least one processor can provide the diagnostic report through a new window (viewport) or a popup.

[0393] In this way, users can retrieve the linked diagnostic report if they are curious about more detailed analysis results while viewing snapshot images. At this time, they can open a new window (or a second viewport) to load the report and display it side-by-side for comparison with the original snapshot.

[0394] Ultimately, the computing device according to the embodiment of the present disclosure stores snapshot information generated during the data screening process and a diagnostic report describing the characteristics of the data set in detail in conjunction, and then cross-references them according to user input, thereby supporting the improvement and utilization of data quality by flexibly switching between intuitive visual information and precise analysis results.

[0396] Although the embodiments have been described above with reference to limited examples and drawings, those skilled in the art can make various modifications and variations from the description above. For example, suitable results can be achieved even if the described techniques are performed in a different order than described, and / or the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0397] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.

Claims

Claim 1 A computing device comprises: a memory; and at least one processor electronically connected to the memory; wherein at least one processor implemented to execute at least one instruction stored in the memory comprises: an operation of acquiring a first data set; an operation of selecting a third level among a plurality of levels classified according to a data processing method based on user input—the plurality of levels including a first level corresponding to a data processing method for diagnosing quantitative indicators of the data set, a second level corresponding to a data processing method for vectorizing and diagnosing the data set using at least one pre-trained artificial intelligence model, and a third level corresponding to a data processing method for diagnosing the data set using an artificial intelligence model built after building an artificial intelligence model optimized for the data set—an operation of determining at least one attribute for a first data processing model corresponding to the third level; and an operation of providing prior information related to the task of processing the first data set using the first data processing model through a first GUI (Graphic User Interface). An operation of constructing a first data processing model including a first artificial intelligence model by obtaining a first artificial intelligence model for analyzing the intrinsic characteristics of the first data set based on user input to the first GUI, wherein the first artificial intelligence model is obtained by fine-tuning at least one foundation model using the first data set; an operation of inputting the first data set into a second data processing model corresponding to the second level, wherein the second data processing model includes a pre-trained second artificial intelligence model, thereby obtaining a first vector set defined in a first-dimensional embedding region from the second data processing model;A computing device configured to perform the operation of inputting the first data set into the first data processing model based on a trigger signal received after the construction of the first data processing model, and obtaining a second vector set defined in a second-dimensional embedding region from the first data processing model; wherein the first dimension in which the first vector set is defined and the second dimension in which the second vector set is defined are different, the first dimension is determined by the second artificial intelligence model, and the second dimension is adaptively determined according to the first data set. Claim 2 delete Claim 3 In claim 1, the computing device comprising at least one of the structure of an artificial intelligence model, parameters of an artificial intelligence model, or hyperparameters of an artificial intelligence model. Claim 4 A computing device according to claim 1, wherein the operation of constructing the first data processing model comprises: the operation of obtaining a plurality of representative values ​​from a plurality of transformation modules based on the first data set; the operation of selecting a first transformation module that satisfies a predetermined condition based on the plurality of representative values; and the operation of constructing the first data processing model by obtaining the first artificial intelligence model including a pre-trained third artificial intelligence model and the first transformation module. Claim 5 A computing device according to claim 1, wherein the operation of constructing the first data processing model comprises: the operation of inputting the first data set into a plurality of foundation models; the operation of obtaining a plurality of representative values ​​based on data output from the plurality of foundation models; the operation of determining a first foundation model that satisfies a predetermined condition based on the plurality of representative values; and the operation of constructing the first data processing model by obtaining the first artificial intelligence model including the first foundation model. Claim 6 A computing device according to claim 1, wherein the prior information comprises at least one of cost information associated with the cost incurred in processing the data set, result preview information predicting and presenting the result of processing the data set, progress information indicating the progress of processing the data set, reference result information indicating the processing result of a case similar to the data set, or attribute information indicating the attribute of a processing model used for data processing. Claim 7 A computing device according to claim 1, wherein the first data processing model comprises at least one of tools storing a processing algorithm for processing the first data set. Claim 8 A computing device according to claim 1, wherein the user input to the first GUI includes an approval input for processing a first data set using the first data processing model. Claim 9 A computing device according to claim 1, further configured to perform: an operation in which at least one processor obtains a first diagnostic result for the first data set by calculating at least one characteristic value of data included in the first data set based on the first vector set; and an operation in which a second diagnostic result for the first data set is obtained by calculating at least one characteristic value of data included in the first data set based on the second vector set. Claim 10 In claim 9, the computing device configured to further perform the operation of providing the first vector set or the second vector set to a first visualization tool to provide a first data image of two dimensions or three dimensions—the first data image corresponds to the first data set. Claim 11 In a data processing method, the operation of acquiring a first data set by at least one processor executing at least one instruction stored in memory; the operation of selecting a third level among a plurality of levels classified according to a data processing method based on user input—the plurality of levels including a first level corresponding to a data processing method for diagnosing quantitative indicators of the data set, a second level corresponding to a data processing method for diagnosing the data set by vectorizing it using at least one pre-trained artificial intelligence model, and a third level corresponding to a data processing method for diagnosing the data set using an artificial intelligence model constructed after constructing an artificial intelligence model optimized for the data set—the operation of determining at least one attribute for a first data processing model corresponding to the third level; the operation of providing prior information related to the task of processing the first data set using the first data processing model through a first GUI (Graphic User Interface); and the operation of a first artificial intelligence model for analyzing the intrinsic characteristics of the first data set based on user input to the first GUI—the first artificial intelligence model using the first data set to form at least one foundation model The operation of constructing the first data processing model including the first artificial intelligence model by obtaining - obtained by fine-tuning; the operation of inputting the first data set into a second data processing model corresponding to the second level - the second data processing model includes a pre-trained second artificial intelligence model - to obtain a first vector set defined in a first-dimensional embedding region from the second data processing model; and the operation of inputting the first data set into the first data processing model based on a trigger signal received after the construction of the first data processing model to obtain a second vector set defined in a second-dimensional embedding region from the first data processing model;A data processing method comprising, wherein the first dimension in which the first vector set is defined and the second dimension in which the second vector set is defined are different, the first dimension is determined by the second artificial intelligence model, and the second dimension is adaptively determined according to the first data set.; Claim 12 In claim 11, the operation of constructing the first data processing model comprises: the operation of obtaining a plurality of representative values ​​from a plurality of transformation modules based on the first data set; the operation of selecting a first transformation module that satisfies a predetermined condition based on the plurality of representative values; and the operation of constructing the first data processing model by obtaining the first artificial intelligence model including a pre-trained third artificial intelligence model and the first transformation module. Claim 13 A data processing method according to claim 11, wherein the above-mentioned prior information comprises at least one of cost information associated with the cost incurred in processing the data set, result preview information that predicts and displays the result of processing the data set in advance, progress information indicating the progress of processing the data set, reference result information indicating the processing result of a case similar to the data set, or attribute information indicating the attribute of the processing model used for data processing. Claim 14 A data processing method according to claim 11, further comprising: an operation of obtaining a first diagnostic result for the first data set by calculating at least one characteristic value of the data included in the first data set based on the first vector set; and an operation of obtaining a second diagnostic result for the first data set by calculating at least one characteristic value of the data included in the first data set based on the second vector set. Claim 15 delete