Artificial intelligence model pipeline automation system and automatic execution method based on user interface

Abstract

The present invention relates to an artificial intelligence model pipeline automation system and an automatic execution method thereof, which enables a domain expert without coding knowledge to control the entire pipeline process from the training to the deployment of an artificial intelligence model solely through intuitive user interface operations. The present invention provides an Abstraction Mapping Layer (AML) between the user interface section and the pipeline execution layer to automatically convert UI actions input by the user into internal pipeline commands, thereby solving the problem of having to manually configure a complex infrastructure environment. Furthermore, based on the converted commands, the offline pipeline execution section can independently and automatically execute pipeline steps, including data preprocessing, incremental fine-tuning, and non-disruptive deployment, within the closed network environment of the edge device using a local model repository and an embedded inference engine, without a connection to an external cloud network. This enables even non-experts to easily train customized AI models and immediately apply them in the field without the need for coding in languages ​​like Python or complex configurations. Furthermore, by eliminating dependence on external clouds, AI models can be safely and flexibly optimized and commercialized without the risk of data leakage, even in strict security environments where data export is restricted.

Description

Technology Field

[0001] The present invention relates to a pipeline control system and a method for executing the same for automating data processing of artificial intelligence models, machine learning, and offline model deployment based on edge computing.

[0002] More specifically, the invention relates to an artificial intelligence model pipeline automation system and an automatic execution method thereof, which includes an Abstraction Mapping Layer (AML) between a user interface section and a pipeline execution layer to convert input UI actions into internal pipeline commands so that even users without coding knowledge can control the entire pipeline process through an intuitive interface, and automatically executes pipeline stages including data preprocessing, incremental fine-tuning, and zero-downtime deployment independently within a closed edge device environment using a local model repository and an embedded inference engine without connection to an external cloud network. Background Technology

[0003] Typically, applying artificial intelligence (AI) models to actual industrial sites requires the establishment of a pipeline that spans from data collection, validation, preprocessing, model training, fine-tuning, and final deployment.

[0004] As the demand for domain-specific AI models has recently surged across various industries, diverse development tools and platforms are being proposed to manage and operate these pipelines.

[0005] However, conventionally, users had to write code directly using specialized programming languages ​​such as Python to execute each stage of the artificial intelligence model pipeline.

[0006] In addition, there is a limitation in that the barrier to entry is very high for field domain experts lacking coding knowledge to directly train and optimize artificial intelligence models, as complex cloud infrastructure environments must be manually configured and parameters tuned.

[0007] Furthermore, conventional AI development and service environments are mostly structured to rely heavily on external cloud servers. As a result, the model training environment and the field deployment environment where actual services run are physically and logically separated, making it difficult to apply changes in the field immediately.

[0008] In particular, in industries with strict data security and personal information protection regulations, such as healthcare, the export of internal data to external cloud networks is often fundamentally prohibited. Consequently, existing cloud-based AI platforms face structural limitations in their application within closed network environments. Furthermore, conventional methods make it difficult to provide flexible responses, such as replacing field agents without service interruption or immediately rolling back versions in the event of issues.

[0009] Therefore, there is an urgent need for technology that allows non-developer users to easily control the entire pipeline, from data preprocessing to model training and deployment, solely through intuitive user interface (UI) operations, without the need for coding or complex configurations.

[0010] Furthermore, there is a need to develop an AI model pipeline automation platform that automatically converts input user actions into internal system commands and can independently complete fine-tuning and non-disruptive deployment within the limited local environment of edge devices without connection to external communication networks. The problem to be solved

[0011] The problem that the present invention aims to solve is to provide an artificial intelligence model pipeline automation system and an automatic execution method thereof, which allows even domain experts without professional coding knowledge to easily control the entire pipeline process, from data preprocessing and incremental fine-tuning to deployment of an artificial intelligence model, solely through intuitive user interface (UI) operations.

[0012] Furthermore, the problem that the present invention aims to solve is to prevent the user from having to manually configure complex cloud infrastructure environments and parameters by providing an Abstraction Mapping Layer (AML) between the user interface section and the pipeline execution layer to automatically convert user input actions into internal pipeline commands.

[0013] Furthermore, the problem that the present invention aims to solve is to provide an offline pipeline automation environment that enables the independent optimization of artificial intelligence models without data leakage by utilizing a local model repository and an embedded inference engine within a closed network environment of an edge device disconnected from an external communication network, in order to resolve data export restrictions and security issues arising from reliance on external cloud servers.

[0014] Furthermore, the problem that the present invention aims to solve is to provide a flexible operational platform that overcomes the conventional limitations where immediate field application was impossible due to the separation of the training environment and the deployment environment, by supporting zero-downtime deployment and immediate version rollback functions that replace domain-specific agents without interrupting the running service when applying a fine-tuned artificial intelligence model. means of solving the problem

[0015] The present invention, for solving the above problem, provides an artificial intelligence model pipeline automation system and an automatic execution method using the same.

[0016] An artificial intelligence model pipeline automation system according to one embodiment of the present invention may include a user interface unit that receives a user interface (UI) action for training and distributing an artificial intelligence model from a user; a mapping control unit that includes an abstraction mapping layer (AML) located between the user interface unit and the pipeline execution layer and converts the received UI action into an internal pipeline command; and an offline pipeline execution unit that automatically executes pipeline steps including data preprocessing, fine-tuning, and distributing the artificial intelligence model independently within an edge device environment using a local model repository and an embedded inference engine without a connection to an external cloud network, based on the converted internal pipeline command.

[0017] The mapping control unit can parse the input UI action into a standardized Intent Object and dynamically route a pipeline sequence to be executed in the offline pipeline execution unit based on the parsed Intent Object.

[0018] The above user interface unit may include an input module that receives UI actions including a first action for specifying training data, a second action for setting conditions, a third action for field deployment, and a fourth action for version rollback, and a control module that transmits a control signal to the offline pipeline execution unit to automatically perform data validation, automatic preprocessing, incremental fine-tuning, and domain-specific agent generation in response to any one of the first to fourth actions.

[0019] The above offline pipeline execution unit may include a learning control module that sets dynamic learning rules based on natural language or a condition builder within the edge device environment, and a deployment management module that performs zero-downtime deployment by replacing domain-specific agents without interrupting the running service when applying the fine-tuned artificial intelligence model.

[0020] At this time, the edge device environment is a closed network environment isolated from an external communication network, and the offline pipeline execution unit calls a base model stored in the local model repository within the closed network environment and can independently optimize the base model on the built-in inference engine by utilizing only the data input through the user interface.

[0021] In addition, an artificial intelligence model pipeline automatic execution method according to one embodiment of the present invention may include a step in which a user interface unit receives a UI (User Interface) action for training and distributing an artificial intelligence model from a user; a step in which a mapping control unit converts the received UI action into an internal pipeline command through an Abstraction Mapping Layer (AML); and a step in which an offline pipeline execution unit automatically executes a pipeline step including data preprocessing, fine-tuning, and distributing of the artificial intelligence model independently within an edge device environment using a local model repository and an embedded inference engine without a connection to an external cloud network, based on the converted internal pipeline command.

[0022] The step of converting into the internal pipeline command may include the step of the mapping control unit parsing the input UI action into a standardized Intent Object, and the step of the mapping control unit dynamically routing a pipeline sequence to be executed in the offline pipeline execution unit based on the parsed Intent Object. Effects of the invention

[0023] According to the present invention, by providing an Abstraction Mapping Layer (AML) between the user interface section and the pipeline execution layer, intuitive UI actions input by the user—such as specifying training data, setting conditions, field deployment, and version rollback—are automatically converted into internal pipeline commands. This provides a technical effect that allows field domain experts to directly and easily control the entire pipeline process, from data preprocessing and fine-tuning to deployment of an artificial intelligence model, without requiring specialized coding knowledge or the configuration of a complex cloud infrastructure environment.

[0024] In addition, according to the present invention, by automatically executing an offline pipeline within an edge device environment using a local model repository and an embedded inference engine without a connection to an external communication network, it is possible to independently and safely optimize a base model without the risk of information leakage, even in closed network environments such as medical, public, and defense sectors where the external export of data is strictly prohibited.

[0025] Furthermore, according to the present invention, the mapping control unit parses the input UI action into a standardized Intent Object to dynamically route the pipeline sequence and supports dynamic learning rule settings based on natural language or a Condition Builder, thereby maximizing user convenience.

[0026] In addition, when applying a fine-tuned AI model to the field, zero-downtime deployment and immediate version rollback functions are performed, allowing domain-specific agents to be replaced in real time without interrupting the running service, thereby increasing system availability and significantly reducing economic costs associated with maintenance.

[0027] Consequently, according to the present invention, the structural limitations of conventional technology, which required programming knowledge such as Python and relied entirely on external cloud servers, are overcome, making immediate field application impossible due to the separation of the learning environment and the deployment environment. Furthermore, it offers the excellent effect of enabling even non-developers without coding knowledge to agilely and flexibly build and commercialize customized artificial intelligence services within a highly secure environment. Brief explanation of the drawing

[0028] FIG. 1 is a block diagram showing the overall configuration of an artificial intelligence model pipeline automation system according to an embodiment of the present invention. FIG. 2 is a block diagram showing the detailed configuration of a user interface unit and an offline pipeline execution unit according to an embodiment of the present invention. FIG. 3 is a conceptual diagram showing a structure for converting UI actions into internal pipeline commands and routing them through an abstraction mapping layer (AML) of a mapping control unit according to an embodiment of the present invention. FIG. 4 is an example diagram showing the process of independent optimization and non-disruptive deployment of an artificial intelligence model within a closed network environment of an edge device disconnected from an external communication network according to an embodiment of the present invention. FIG. 5 is a flowchart showing an overall method for automatically executing an artificial intelligence model pipeline according to an embodiment of the present invention. FIG. 6 is a flowchart showing the detailed process of the step of converting into internal pipeline commands of FIG. 5 according to an embodiment of the present invention. Specific details for implementing the invention

[0029] Hereinafter, various embodiments of the present invention will be described in detail with reference to the attached drawings so that those skilled in the art can easily implement the present invention. The present invention may be embodied in various different forms and is not limited to the embodiments described herein.

[0030] To clearly explain the present invention, parts unrelated to the explanation have been omitted, and the same reference numerals are assigned to identical or similar components throughout the specification. Accordingly, the reference numerals described above may also be used in other drawings.

[0031] Furthermore, the size and thickness of each component shown in the drawings are depicted arbitrarily for convenience of explanation, and thus the present invention is not necessarily limited to what is illustrated. Thickness may be exaggerated in the drawings to clearly represent various layers and regions.

[0032] Furthermore, the expression "identical" in the explanation may mean "substantially identical." In other words, it may be an identicality to the extent that a person with ordinary knowledge would accept it as identical. Other expressions may also be those in which "substantially" has been omitted.

[0033] Additionally, when a part in the description is described as 'comprising' a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. As used in this specification, 'part' refers to a unit that processes at least one function or operation, and may mean, for example, software, FPGA, or hardware components. The function provided by the 'part' may be performed separately by multiple components or integrated with other additional components. The 'part' in this specification is not necessarily limited to software or hardware, and may be configured to reside in an addressable storage medium or configured to run one or more processors.

[0034] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0036] FIG. 1 is a block diagram showing the overall configuration of an artificial intelligence model pipeline automation system according to one embodiment of the present invention.

[0037] Referring to FIG. 1, the artificial intelligence model pipeline automation system (10) according to the present invention is an integrated platform that enables non-developers without coding knowledge (e.g., field domain experts) to train and deploy artificial intelligence models with only intuitive operations.

[0038] The above artificial intelligence model pipeline automation system (10) is configured to include a user interface unit (100), a mapping control unit (200), a pipeline execution layer (400), and an offline pipeline execution unit (300), and is characterized by having a local model repository (500) and an embedded inference engine (600) to operate independently within an edge device environment (700) isolated from an external communication network.

[0040] FIG. 2 is a block diagram showing the detailed configuration of a user interface unit and an offline pipeline execution unit according to an embodiment of the present invention.

[0041] Referring to FIG. 1 and FIG. 2 together, the user interface unit (100) can perform the role of receiving UI (User Interface) actions for training and distributing an artificial intelligence model from a user. To this end, the user interface unit (100) may include an input module (110) and a control module (120).

[0042] The above input module (110) can provide an intuitive graphical user interface (GUI) environment, such as a touchscreen, mouse click, and drag and drop.

[0043] Specifically, the input module (110) can receive UI actions including a first action for specifying training data (e.g., clicking the "LEARN THIS" button), a second action for setting conditions (e.g., toggling "SET RULE"), a third action for field deployment (e.g., dragging and dropping "DEPLOY"), and a fourth action for rolling back versions (e.g., clicking the "ROLLBACK" button).

[0044] The control module (120) can generate and transmit control signals to the offline pipeline execution unit (300) to automatically perform data validation, automatic preprocessing, progressive fine-tuning, and the creation of a domain-specific agent (810) in response to any one of the first to fourth actions input through the input module (110).

[0046] FIG. 3 is a conceptual diagram showing a structure that converts UI actions into internal pipeline commands and routes them through an abstract mapping layer (AML) of a mapping control unit according to an embodiment of the present invention.

[0047] As illustrated in FIG. 3, the mapping control unit (200) is located between the user interface unit (100) and the pipeline execution layer (400) where the actual script is executed, and can perform the role of a kind of interpreter.

[0048] The mapping control unit (200) may include an abstraction mapping layer (210, AML: Abstraction Mapping Layer).

[0049] Previously, complex Python-based code had to be written, but the abstraction mapping layer (210) can automatically convert simple UI actions of the user into complex internal pipeline commands required by the backend.

[0050] More specifically, when the mapping control unit (200) receives a user's UI action, it can parse it into a standardized intent object (Intent Object, e.g., JSON format {"action": "train", "dataset_path": " / local / data", "hyperparameters": "auto"}).

[0051] Afterwards, based on the parsed intent object, the pipeline sequence to be executed currently among various modules in the offline pipeline execution unit (300) can be dynamically routed.

[0052] For example, if the first action is parsed as an intent object, it can be routed to the data preprocessing and fine-tuning sequence, and if the third action is parsed, it can be routed to the deployment sequence.

[0054] FIG. 4 is an exemplary diagram illustrating the process of independent optimization and uninterrupted deployment of an artificial intelligence model within a closed network environment of an edge device disconnected from an external communication network, according to one embodiment of the present invention.

[0055] Referring to FIG. 2 and FIG. 4 together, the offline pipeline execution unit (300) can perform the role of actually automatically executing pipeline stages (data preprocessing, fine-tuning, distribution, etc.) based on internal pipeline commands converted through the mapping control unit (200).

[0056] The offline pipeline execution unit (300) may include a learning control module (310) and a distribution management module (320).

[0057] The above learning control module (310) can support setting dynamic learning rules based on a natural language input window or a condition builder in the form of visual blocks. This allows even users who do not know coding to easily create condition-based rules such as "start learning when defective product data exceeds 50."

[0058] The above distribution management module (320) can perform a zero-downtime deployment to replace the domain-specific agent (810) without interrupting the currently running service when applying the fine-tuned model to the actual field.

[0059] For example, through a Blue-Green deployment method or a method of instantaneously switching inference pointers in memory, it is possible to implement real-time intelligence updates without interrupting operations of factory lines or hospital medical devices.

[0060] At this time, the most important feature of the present invention is that the edge device environment (700) is an air-gapped environment that is completely disconnected from the external internet.

[0061] Unlike conventional cloud methods, the offline pipeline execution unit (300) can exclusively use only the local model repository (500) and the GPU / NPU-based built-in inference engine (600) provided inside the edge device without connecting to an external cloud network such as AWS or GCP.

[0062] The offline pipeline execution unit (300) can call a pre-trained base model (800) stored in a local model repository (500) and fine-tune the base model (800) on the built-in inference engine (600) using only internal local data input through the user interface unit (100).

[0063] The resulting domain-specific agent (810) can also be immediately applied to internal services without being exported externally. Through this structure, zero risk of data leakage can be achieved in fields where security is critical, such as military secrets, patient medical information, and core manufacturing data.

[0065] FIG. 5 is a flowchart showing an overall method for automatically executing an artificial intelligence model pipeline according to an embodiment of the present invention, and FIG. 6 is a flowchart showing a detailed process of the step of converting into an internal pipeline instruction of FIG. 5.

[0066] Referring to FIGS. 5 and FIGS. 6, the time-series flow of the automatic execution method performed in the system (10) is described.

[0067] First, the user interface unit (100) can receive UI actions for training and distributing the artificial intelligence model from the user (e.g., specifying training data, setting conditions, field distribution, version rollback, etc.).

[0068] Next, the mapping control unit (200) can convert the input UI action into an internal pipeline instruction that the computer can understand through the abstraction mapping layer (210).

[0069] Referring to FIG. 6, this step may specifically include the step of the mapping control unit (200) parsing the input UI action into a standardized intent object, and the step of dynamically routing an appropriate pipeline sequence to be executed in the offline pipeline execution unit (300) based on the attribute value of the parsed intent object.

[0070] Finally, the offline pipeline execution unit (300) can automatically execute the pipeline within the edge device environment (700) based on the converted internal pipeline command and routing information.

[0071] At this time, the entire process of optimizing the base model (800) and deploying it to a domain-specific agent (810) without interruption can be performed independently by utilizing only the internal local model repository (500) and the built-in inference engine (600) in a limited manner, without intervening in communication with an external cloud network.

[0073] The drawings and detailed description of the invention referenced so far are merely exemplary of the invention and are used only for the purpose of explaining the invention, not to limit the meaning or the scope of the invention as defined in the claims. Therefore, those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of the invention should be determined by the technical spirit of the appended claims.

[0074] The embodiments described above may be implemented as hardware components, software components, and / or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may be implemented using one or more general-purpose computers or special-purpose computers, such as, for example, a processor, a controller, an Arithmetic Logic Unit (ALU), a Digital Signal Processor (DSP), a microcomputer, a Field Programmable Gate Array (FPGA), a Programmable Logic Unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions.

[0075] The processing unit may execute an operating system and one or more software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of understanding, the processing unit may be described as being used as a single unit, but a person of ordinary skill in the art will understand that the processing unit may include multiple processing elements and / or multiple types of processing elements.

[0076] For example, the processing unit may include multiple processors or one processor and one controller. Additionally, other processing configurations, such as a parallel processor, are also possible. Software may include a computer program, code, instructions, or a combination of one or more of these, and may configure the processing unit to operate as desired or command the processing unit independently or collectively.

[0077] Software and / or data may be embodied in any type of machine, component, physical device, virtual equipment, computer storage medium, or device so as to be interpreted by a processing device or to provide instructions or data to a processing device. Software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

[0078] The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiment, or they may be those known and available to those skilled in the art of computer software.

[0079] Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operation of the embodiments, and vice versa.

[0080] Although the embodiments have been described above with reference to limited embodiments and drawings, those skilled in the art can make various modifications and variations from the description above. For example, appropriate results may be achieved even if the described techniques are performed in a different order than described, and / or if 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. Therefore, other implementations, other embodiments, and equivalents to the claims are also included within the scope of the claims set forth below. Explanation of the symbols

[0081] 10: Artificial Intelligence Model Pipeline Automation System 100: User Interface Section 110: Input module 120: Control Module 200: Mapping control unit 210: Abstraction Mapping Layer (AML) 300: Offline pipeline execution part 310: Learning Control Module 320: Deployment Management Module 400: Pipeline Execution Layer 500: Local model repository 600: Built-in inference engine 700: Edge device environment (closed network environment) 800: Base model 810: Domain-Specific Agent

Claims

Claim 1 An artificial intelligence model pipeline automation system comprises: a user interface unit that receives UI (User Interface) actions for training and deploying an artificial intelligence model from a user; a mapping control unit located between the user interface unit and the pipeline execution layer and including an Abstraction Mapping Layer (AML) that converts the received UI actions into internal pipeline commands; and an offline pipeline execution unit that automatically executes pipeline steps including data preprocessing, fine-tuning, and deployment of the artificial intelligence model independently within an edge device environment using a local model repository and an embedded inference engine without a connection to an external cloud network, based on the converted internal pipeline commands; wherein the mapping control unit parses the received UI actions into standardized Intent Objects and dynamically routes a pipeline sequence to be performed in the offline pipeline execution unit based on the parsed Intent Objects. Claim 2 delete Claim 3 An artificial intelligence model pipeline automation system according to claim 1, wherein the user interface unit comprises: an input module that receives UI actions including a first action for designating training data, a second action for setting conditions, a third action for field deployment, and a fourth action for version rollback; and a control module that transmits a control signal to the offline pipeline execution unit to automatically perform data validation, automatic preprocessing, incremental fine-tuning, and domain-specific agent generation in response to any one of the first to fourth actions. Claim 4 An artificial intelligence model pipeline automation system according to claim 1, wherein the offline pipeline execution unit comprises: a learning control module that sets dynamic learning rules based on natural language or a condition builder within the edge device environment; and a deployment management module that performs zero-downtime deployment by replacing a domain-specific agent without interrupting a running service when applying the fine-tuned artificial intelligence model. Claim 5 An artificial intelligence model pipeline automation system according to claim 1, wherein the edge device environment is a closed network environment isolated from an external communication network, and the offline pipeline execution unit calls a base model already stored in the local model repository within the closed network environment and independently optimizes the base model on the built-in inference engine by utilizing only the data input through the user interface unit. Claim 6 A method for automatically executing a pipeline in an artificial intelligence model pipeline automation system comprises: a step in which a user interface unit receives a UI (User Interface) action for training and deploying an artificial intelligence model from a user; a step in which a mapping control unit converts the received UI action into an internal pipeline command through an Abstraction Mapping Layer (AML); and a step in which an offline pipeline execution unit, based on the converted internal pipeline command, automatically executes a pipeline step including data preprocessing, fine-tuning, and deployment of the artificial intelligence model independently within an edge device environment using a local model repository and an embedded inference engine without a connection to an external cloud network; wherein the step of converting into an internal pipeline command comprises: a step in which the mapping control unit parses the received UI action into a standardized Intent Object; and a step in which the mapping control unit dynamically routes a pipeline sequence to be executed in the offline pipeline execution unit based on the parsed Intent Object.

Images