KR20220075965A - Device and method for supporting deep learning - Google Patents

Abstract

A deep learning support device and a deep learning support method are provided, and according to an embodiment, as a support device communicating with an edge device performing deep learning, a memory in which a program for performing deep learning is stored, and executing the program and a controller for performing deep learning by doing so, wherein the controller may provide an edge-supported deep learning model learned based on the learning data obtained from the edge device to the edge device.

Description

DEVICE AND METHOD FOR SUPPORTING DEEP LEARNING

Embodiments disclosed herein relate to a device and a method for supporting deep learning, and more particularly, to a device and method for supporting performing deep learning in an edge device.

Recently, with the development of computing technology, the application of machine learning is increasing, and among recent machine learning, deep learning technology is developing rapidly.

Meanwhile, edge devices in which computing is performed at the location of users or data have become widespread. As an example, a camera implemented as an edge device is installed in an industrial site such as a factory to perform machine vision, and as another example, a camera implemented as an edge device is installed on a road such as a traffic intersection or alley to reduce traffic. It also controls or manages compliance with traffic laws.

Since these edge devices process a lot of data, attempts are being made to apply a deep learning model to improve its performance.

However, since edge devices are installed and used at the user end, they operate within a limited environment and their size and performance are often reduced accordingly. , in order to apply a deep learning model to an edge device, it is necessary to separately train the deep learning model stored in the edge device, and there is the inconvenience of separately collecting training data for learning the deep learning model.

Therefore, there are many difficulties in implementing an edge device that can perform deep learning.

In relation to this, a prior art document, Republic of Korea Patent Registration No. 10-2016-0096460, relates to a deep learning-based recognition system, wherein the system performs learning on an input image according to a first classification criterion to output a first classification result. a first classifier and a second classifier configured to output a second classification result by performing learning on the first classification result according to a second classification criterion different from the first classification criterion, wherein the first classification criterion is the second classification criterion It describes what is in the criteria that is a prerequisite for That is, the prior art does not propose a method for realizing an edge device that performs deep learning only by outputting a result for an input value based on deep learning.

Therefore, there is a need for a technique for solving the above-mentioned problems.

On the other hand, the above-mentioned background art is technical information that the inventor possessed for the purpose of derivation of the present invention or acquired during the derivation process of the present invention, and it cannot be said that it is necessarily known technology disclosed to the general public before the filing of the present invention. .

Embodiments disclosed herein, it is an object to present a deep learning support device and deep learning support method.

In addition, the embodiments disclosed herein have an object to provide a deep learning support device and deep learning support method that can support the deep learning performance of the edge device with a minimum amount of time and minimum resources.

In addition, the embodiments disclosed in the present specification, it is an object to present a deep learning support device and deep learning support method that can collect learning data for learning of the edge device at the same time as installation.

In addition, embodiments disclosed herein have an object to provide a deep learning support device and a deep learning support method that do not require real-time communication between the edge device and the support device.

As a technical means for achieving the above-described technical problem, according to an embodiment described in the present specification, as a support device that communicates with an edge device that performs deep learning, a memory in which a program for performing deep learning is stored, and the and a controller for performing deep learning by executing a program, wherein the controller may provide an edge-supported deep learning model learned based on the learning data obtained from the edge device to the edge device.

In addition, as a technical means for achieving the above-described technical problem, according to an embodiment described in the present specification, a support device that communicates with an edge device performing deep learning, a method for supporting the deep learning performance of the edge device As such, it may include acquiring training data from the edge device, and providing an edge-supported deep learning model learned based on the acquired training data to the edge device.

In addition, as a technical means for achieving the above-described technical problem, according to an embodiment described in the present specification, as a computer-readable recording medium on which a program for performing a support method is recorded, the support method includes learning data from an edge device. It may include obtaining, and providing an edge-supported deep learning model learned based on the obtained learning data to the edge device.

In addition, as a technical means for achieving the above-described technical problem, according to an embodiment described in the present specification, it is performed by a support device that communicates with an edge device that performs deep learning, and is stored in the medium to perform the support method. As a computer program, the support method may include acquiring training data from an edge device, and providing an edge-supported deep learning model learned based on the acquired training data to the edge device.

In addition, as a technical means for achieving the above-described technical problem, according to an embodiment described in the present specification, a deep learning system including an edge device performing deep learning, and a support device communicating with the edge device, the A method for supporting the deep learning performance of an edge device, the support device acquiring learning data from the edge device, the support device, the edge support deep learning model learned based on the acquired learning data It may include providing the edge device to the edge device, and the edge device configuring an edge-only deep learning model based on the edge supported deep learning model.

According to one of the above-mentioned problem solving means, it is possible to present a deep learning support device and a deep learning support method.

In addition, according to one of the above-mentioned problem solving means, it is possible to present a deep learning support device and a deep learning support method that can support the deep learning performance of an edge device with a minimum amount of time and minimum resources. In addition, learning can proceed without costly and time-consuming manual labeling every time an edge device is installed.

In addition, according to one of the above-described problem solving means, it is possible to present a deep learning support device and a deep learning support method that can collect learning data for learning of an edge device at the same time as installation. Through this, there is no need to separately collect learning data by investing extra time or resources to implement deep learning of edge devices.

In addition, according to one of the above-described problem solving means, it is possible to present a deep learning support device and a deep learning support method that do not require real-time communication between the edge device and the support device. That is, since real-time inference is performed by the edge device, real-time communication between the edge device and the supported device is not required, so the bandwidth for communication between the two devices does not need to be wide.

Effects obtainable in the disclosed embodiments are not limited to the above-mentioned effects, and other effects not mentioned are clear to those of ordinary skill in the art to which the embodiments disclosed from the description below belong. can be understood clearly.

1 is a configuration diagram for explaining a deep learning system according to an embodiment disclosed in the present specification.
2 is a block diagram showing the configuration of a deep learning support device according to an embodiment disclosed in the present specification.
3 to 4 are flowcharts for explaining a deep learning support method according to an embodiment described in the present specification.

Hereinafter, various embodiments will be described in detail with reference to the accompanying drawings. The embodiments described below may be modified and implemented in various different forms. In order to more clearly describe the characteristics of the embodiments, detailed descriptions of matters widely known to those of ordinary skill in the art to which the following embodiments belong are omitted. In addition, in the drawings, parts irrelevant to the description of the embodiments are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a component is said to be "connected" with another component, it includes not only the case where it is 'directly connected', but also the case where it is 'connected with another component in between'. In addition, when a component "includes" a component, it means that other components may be further included, rather than excluding other components, unless otherwise stated.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

1 is a configuration diagram for explaining a deep learning system according to an embodiment disclosed in the present specification.

1, the deep learning system 100 includes an edge device 10 and a deep learning support device 20, and the edge device 10 and the deep learning support device 20 communicate through a network. can

The edge device 10 is implemented to be lightweight compared to the deep learning support device 20, and accordingly, for example, FLOPS (FLoating point Operations Per Second), RAM (RAM) size, GPU size, image resolution and processing speed (fps) at least one may have a relatively lower performance compared to the deep learning support device (20).

In relation to this, it is possible to modify a lightweight network (mobilenet-type network) that can operate at 30 FPS (real-time basis) or more in commercially available deep learning equipment (intel NCS2, NVIDIA Jetson NANO, etc.) and use it in the edge device 10 .

The edge device 10 is a device that performs deep learning on input data and outputs a result, and a program for performing a deep learning model can be installed and driven, and the edge device 10 is a computer program that is driven The result can be provided by performing deep learning on the input data according to the control.

The edge device 10 may store an edge-only deep learning model as a deep learning model, and the edge-only deep learning model is, for example, a deep neural network (DNN), a convolutional neural network (CNN), or a recurrent neural network (RNN). etc. can be implemented. Therefore, when the edge device 10 obtains the data to be processed, it may output a result value based on the learned deep learning model.

As such, the edge device 10 included in the deep learning system 100 performs deep learning on the acquired data and outputs a result value, so there is no need to communicate with the deep learning support device 20 in real time. The bandwidth required does not have to be wide.

Meanwhile, FIG. 2 is a diagram illustrating the configuration of a device for supporting deep learning according to an embodiment.

According to an embodiment, a program for performing a deep learning model may be installed and driven in the support device 20, and the support device 20 according to the embodiment described herein according to the control of the driven computer program. support method can be carried out.

Such a support device 20 may be implemented as a server or a server-client system, and if the support device 20 is implemented as a server-client system and includes an electronic terminal, the electronic terminal is provided from the edge device 10 An interface for receiving the acquired data may be included.

In this case, the electronic terminal may be implemented as a computer, a portable terminal, a television, a wearable device, etc. that can connect to a remote server through a network or connect to other devices and servers. Here, the computer includes, for example, a laptop, a desktop, and a laptop equipped with a web browser, and the portable terminal is, for example, a wireless communication device that ensures portability and mobility. , PCS (Personal Communication System), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), GSM (Global System for Mobile communications), IMT (International Mobile Telecommunication)-2000, CDMA (Code) All kinds of handhelds such as Division Multiple Access)-2000, W-Code Division Multiple Access (W-CDMA), Wireless Broadband Internet (Wibro), Smart Phone, Mobile Worldwide Interoperability for Microwave Access (WiMAX), etc. It may include a (Handheld)-based wireless communication device. In addition, the television may include IPTV (Internet Protocol Television), Internet TV (Internet Television), terrestrial TV, cable TV, and the like. Furthermore, a wearable device is, for example, a type of information processing device that can be worn directly on the human body, such as watches, glasses, accessories, clothes, shoes, etc. can be connected with

In addition, the server may be implemented as a computer capable of communication through a network or may be implemented as a cloud computing server. In addition, the server may include a storage device capable of storing data or may store data through a third server.

As described above, the support device 20 may be implemented in any one form of a server or a server-client system, and the components constituting the support device 20 may be performed in a plurality of physically separated servers or one server. can be performed in

The support device 20 according to the embodiment may include a communication unit 210 , a memory 220 , and a control unit 230 .

The communication unit 210 may allow the support device 20 to perform wired/wireless communication with the edge device 10 or a network. For example, the communication unit 210 may communicate with a third server to transmit/receive various data for performing a support method. To this end, the communication unit 210 may include a communication module that supports at least one of various wired and wireless communication methods. For example, the communication module may be implemented in the form of a chipset.

In this case, the wireless communication supported by the communication unit 210 may be, for example, Wireless Fidelity (Wi-Fi), Wi-Fi Direct, Bluetooth, Ultra Wide Band (UWB), or Near Field Communication (NFC). . In addition, the wired communication supported by the communication unit 210 may be, for example, USB or High Definition Multimedia Interface (HDMI).

According to the embodiment, when the support device 20 is configured as a server, the communication unit 210 may receive, from the edge device 10 , model information or performance information about an edge-only deep learning model stored in the edge device. Also, an operation result according to the execution of the support method may be transmitted to the edge device 10 .

On the other hand, various types of data such as files, applications, and programs may be installed and stored in the memory 220 . The controller 230 may access and use data stored in the memory 220 , or may store new data in the memory 220 . Also, the controller 230 may execute a program installed in the memory 220 . For example, in the memory 220 , a program in which a deep learning model is implemented may be installed, and data obtained from the edge device 10 may be stored as described above.

On the other hand, the controller 230 controls the overall operation of the support device 20 , and may include a processor such as a CPU or GPU. The controller 230 may control other components included in the support device 20 to perform support for performing deep learning. For example, the controller 230 may execute a program stored in the memory 220 , read a file stored in the memory 220 , or store a new file in the memory 220 .

According to an embodiment, the controller 230 may learn a labeling network for labeling data obtained from the edge device 10 .

In this case, the deep learning model applied to the labeling network may be, for example, a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), and the like, and is not limited to the above-described example.

According to another embodiment, the labeling network is stored in the controller 230 as a general-purpose labeling network, so learning is not required when first labeling the learning data obtained from the edge device 10.

The controller 230 may label data obtained from the edge device 10 using a labeling network.

In addition, the control unit 230 may train the edge-supported deep learning model to select data with high reliability from the labeled data and output the output value of the edge-only deep learning model by using the selected data as an input value.

In addition, the control unit 230 may select data with high reliability from among the labeled data and retrain the labeling network with the selected data.

In this case, the reliability can be increased as the labeling becomes more suitable. For example, when the class of the target image is inferred and labeled with the class to which the target image belongs, the corresponding class-related score can be determined as the reliability.

The controller 230 may provide the learned edge support deep learning model to the edge device 10 .

According to an embodiment, the controller 230 may provide a weight value in the learned edge support deep learning model to the edge device 10 . Accordingly, the edge device 10 receiving the weight value may optimize the edge-only deep learning model by changing the weight value of the edge-only deep learning model according to the weight value received from the support device 20 .

Meanwhile, the controller 230 may obtain model information about the edge-only deep learning model from the edge device 10 for learning the edge-supported deep learning model.

For example, the controller 230 may select a model matching an edge-only deep learning model from among a plurality of deep learning models. For example, when model information indicating that an edge-only deep learning model is implemented as a CNN is obtained from the edge device 10, a CNN matching the corresponding model information can be selected. Also, for example, when the control unit 230 obtains model information including information on the layers constituting the edge-only deep learning model, the layer of the edge-supported deep learning model can be equally configured based on the model information.

In addition, the controller 230 - may obtain performance information about the performance of a device implemented as the edge device 10 from the edge device 10 for learning the edge-supported deep learning model.

For example, the control unit 230 may provide at least one of FLOPS (FLoating point Operations Per Second), RAM size, GPU size, image resolution and processing speed (or inspection speed; FPS) as performance information to the edge device. can be obtained from

In addition, the controller 230 may configure an edge-supported deep learning model based on the model information obtained from the edge device 10 .

For example, when it obtains that the processing speed is a predetermined value, the control unit 230 configures the edge support deep learning model so that when an input value for the edge support deep learning model is input, the output value can be output according to the processing speed. can Alternatively, for example, when it is obtained that the size of the GPU is a predetermined size, the controller 230 may adjust the size of the GPU allocated for the edge-supported deep learning model according to the obtained performance information.

The deep learning system 100 as described above may be implemented in, for example, a machine vision system (not shown).

Here, machine vision means to automate the industry through a camera (visual recognition), CPU, and SW, instead of the existing method that humans judged with the naked eye to inspect or measure an object.

For example, the deep learning system 100 is included in a machine vision system (not shown), includes a machine vision system (not shown), or while communicating with a machine vision system (not shown), determining whether or not a product is acceptable in a smart factory. can be performed.

Hereinafter, the deep learning system 100 will be described as an example included in the machine vision system (not shown).

That is, the machine vision system (not shown) may include the deep learning system 100 according to an embodiment disclosed herein, and accordingly, the edge device 10 and the support device 20 .

Therefore, the product can be photographed through the edge device 10, and the edge device 10 analyzes the projected image based on the edge-only deep learning model implemented by the support device 20 to determine whether the product is normal or defective. or the type of defect can be identified. In addition, it is possible to notify the presence of a defect through the manager terminal, or notify the manager terminal that a defective or normal image of an unlearned type has been received.

To this end, the machine vision system (not shown) may provide the image of the product photographed by the edge device 10 to the support device 20 to label it and train the deep learning model accordingly.

This eliminates the need to perform unnecessary actions such as stopping the line from running in order to secure unexpected learning data from the factory line, and accordingly, a smart factory that can run the line in real time can be implemented.

Meanwhile, FIG. 3 is a flowchart for explaining a method in which the deep learning support device according to an embodiment supports performing deep learning, and FIG. 4 is a method for the deep learning system according to an embodiment to support performing deep learning. is a flowchart for

The support method shown in FIGS. 3 and 4 includes time-series processing in each of the deep learning system 100 and the support device 20 described with reference to FIGS. 1 and 2 . Therefore, even if omitted below, the content described above with respect to the deep learning system 100 and the support device 20 shown in FIGS. 1 and 2 is a support method according to the embodiment shown in FIGS. 3 to 4 . can also be used for

Referring to FIG. 3 , the support device 20 may configure an edge-supported deep learning model (S310), and when acquiring training data, labels the acquired training data based on the labeling network (S320), and has high reliability An edge-supported deep learning model can be trained using the labeled training data as an input (S330).

To this end, the support device 20 obtains model information about the edge-only deep learning model stored in the edge device 10, from the edge device 10, and configures the edge support deep learning model based on the model information. have.

Alternatively, the support device 20 may provide at least one of FLOPS (FLoating point Operations Per Second), RAM size, GPU size, image resolution and processing speed as performance information regarding the edge device 10 to the edge device. It is possible to construct the edge-supported deep learning model by obtaining from (10).

That is, as shown in FIG. 4 , the support device 20 may obtain at least one of model information and performance information from the edge device 10 ( S410 ).

And based on the obtained information, the support device 20 may implement an edge support deep learning model (S310).

Thereafter, the edge device 10 may receive the learning data and may provide the received learning data to the support device 20 ( S420 ).

In this case, the edge device 10 may separately receive learning data for learning, and may also receive the learning data by acquiring a plurality of images to be analyzed in an environment in which the edge device 10 is installed.

Learning data obtained from the edge device 10 may be labeled based on a labeling network. In this case, the edge-supported deep learning model can be trained by using the highly reliable labeled training data as an input value (S330).

And the support device 20 may provide the learned edge support deep learning model to the edge device (S430).

For example, the support device 20 may provide a weight value of the learned edge support deep learning model to the edge device 10 .

As such, the edge device 10 provided with the edge support deep learning model learned from the support device 20 may learn the edge-only deep learning model (S440).

For example, the edge device 10 may configure an edge-only deep learning model by applying a weight value of the learned edge-supported deep learning model to the edge-only deep learning model. In this way, the edge device 10 receives the learned edge support deep learning model from the support device 20 , so that it is possible to configure a trained edge-only deep learning model without the need to learn within the edge device 10 .

And the edge device 10 may process an input value based on the learned edge-only deep learning model and output a result value (S450).

For example, if the edge device 10 is configured with a machine vision system, it may be determined whether a photographed image is an image of a defective product in determining whether the device is defective or not.

Meanwhile, the edge device 10 may determine whether to update the edge-only deep learning model whenever an input value is obtained ( S460 ).

For example, when it is decided to update every predetermined period, the edge device 10 may determine to update the edge-only deep learning model whenever the period approaches. Or, for example, when the number of accumulated input values after the update of the edge-only deep learning model is greater than or equal to a predetermined value, it may be determined that the edge-only deep learning model needs to be updated. Alternatively, if the size of the input value that comes in at once is greater than or equal to a predetermined value, it may be determined that the edge-only deep learning model needs to be updated.

The edge device 10 may perform step S450 until the edge-only deep learning model needs to be updated, and if the edge-only deep learning model needs to be updated, go to step S420 to perform the edge-only deep learning model update. An input value may be provided to the support device 20 . In this case, the accumulated input value may be provided, for example, an input value accumulated since the most recent update of the edge-only deep learning model, an input value accumulated for a predetermined period, etc. may be provided.

According to the embodiments described above, it is possible to implement deep learning with the edge device 10 having a lightweight performance. Accordingly, as the environment in which the edge device 10 is installed is changed, even in a situation where the deep learning model needs to be separately learned, it is possible to learn quickly with minimal resources through the support device 20 .

In addition, for example, in determining whether or not a machine vision system is good or bad, the edge device 10 may detect the presence of the corresponding image through the support device 20 even if a different type of bad image that has not been learned is received.

The deep learning support method described above may also be implemented in the form of a computer-readable medium for storing instructions and data executable by a computer. In this case, the instructions and data may be stored in the form of program code, and when executed by the processor, a predetermined program module may be generated to perform a predetermined operation. In addition, computer-readable media can be any available media that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable medium may be a computer recording medium, which is a volatile and non-volatile and non-volatile storage medium implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. It may include both volatile, removable and non-removable media. For example, the computer recording medium may be a magnetic storage medium such as HDD and SSD, an optical recording medium such as CD, DVD, and Blu-ray disc, or a memory included in a server accessible through a network.

The deep learning support method described above may be implemented as a computer program (or computer program product) including instructions executable by a computer. A computer program includes programmable machine instructions that are processed by a processor, and includes a high-level programming language (RAM) ming language, an object-oriented programming language (RAM) ming language, and assembly It may be implemented in a language or a machine language. In addition, the computer program may be recorded in a tangible computer-readable recording medium (eg, a memory, a hard disk, a magnetic/optical medium, or a solid-state drive (SSD), etc.).

The deep learning support method described above may be implemented by executing the computer program as described above by a computing device. The computing device may include at least a portion of a processor, a memory, a storage device, a high-speed interface connected to the memory and the high-speed expansion port, and a low-speed interface connected to the low-speed bus and the storage device. Each of these components is connected to each other using various buses, and may be mounted on a common motherboard or in any other suitable manner.

Here, the processor may process a command within the computing device, such as, for example, to display graphic information for providing a graphic user interface (GUI) on an external input or output device, such as a display connected to a high-speed interface. Examples are instructions stored in memory or a storage device. In other embodiments, multiple processors and/or multiple buses may be used with multiple memories and types of memory as appropriate. In addition, the processor may be implemented as a chipset formed by chips including a plurality of independent analog and/or digital processors.

Memory also stores information within the computing device. As an example, the memory may be configured as a volatile memory unit or a set thereof. As another example, the memory may be configured as a non-volatile memory unit or a set thereof. The memory may also be another form of computer readable medium, such as, for example, a magnetic or optical disk.

In addition, the storage device may provide a large-capacity storage space to the computing device. A storage device may be a computer-readable medium or a component comprising such a medium, and may include, for example, devices or other components within a storage area network (SAN), a floppy disk device, a hard disk device, an optical disk device, or a tape device, a flash memory, or other semiconductor memory device or device array similar thereto.

The term '~ unit' used in the above embodiments means a hardware component such as software or an FPGA (field program (RAM)mable gate array) or ASIC, and '~ unit' performs certain roles. However, '-part' is not limited to software or hardware. '~' may be configured to reside on an addressable storage medium or may be configured to refresh one or more processors. Accordingly, as an example, '~' indicates components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, and procedures. , subroutines, segments of program patent code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables.

The functions provided in the components and '~ units' may be combined into a smaller number of elements and '~ units' or separated from additional components and '~ units'.

In addition, components and '~ units' may be implemented to play one or more CPUs in a device or secure multimedia card. The above-described embodiments are for illustration, and those of ordinary skill in the art to which the above-described embodiments pertain can easily transform into other specific forms without changing the technical idea or essential features of the above-described embodiments. You will understand. Therefore, it should be understood that the above-described embodiments are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may also be implemented in a combined form.

The scope to be protected through the present specification is indicated by the claims described below rather than the above detailed description, and should be construed to include all changes or modifications derived from the meaning and scope of the claims and their equivalents. .

100: deep learning system
10: Edge device 20: Supported device
210: input/output unit 220: memory
220: communication unit 230: control unit

Claims (12)

As a supporting device that communicates with an edge device performing deep learning,
a memory in which a program for performing deep learning is stored; and
A control unit for performing deep learning by executing the program,
The control unit is
A support device that provides an edge support deep learning model learned based on the learning data obtained from the edge device to the edge device. According to claim 1,
The control unit is
A support device that performs labeling on the learning data obtained from the edge device, learns the edge-supported deep learning model based on the labeled learning data, and provides the learned edge-supported deep learning model to the edge device. According to claim 1,
The control unit is
A support device that acquires model information about an edge-only deep learning model stored in the edge device from the edge device and configures the edge-supported deep learning model based on the model information. According to claim 1,
The control unit is
As performance information about the edge device, at least one of FLOPS (FLoating point Operations Per Second), RAM size, GPU size, image resolution, and processing speed is obtained from the edge device, and based on the performance information A support device constituting the edge-supported deep learning model. An edge device in communication with the support device of claim 1, comprising:
An edge device that configures an edge-only deep learning model based on the edge-supported deep learning model. A support device communicating with an edge device performing deep learning, as a method for supporting deep learning performance of the edge device,
obtaining learning data from the edge device; and
A support method comprising the step of providing an edge support deep learning model learned on the basis of the acquired learning data to the edge device. 7. The method of claim 6,
The step of providing to the edge device comprises:
performing labeling on the learning data obtained from the edge device;
training the edge-supported deep learning model based on the labeled training data; and
A support method comprising the step of providing a learned edge-supported deep learning model to the edge device. 7. The method of claim 6,
acquiring model information about an edge-only deep learning model stored in the edge device from the edge device; and
Further comprising the step of configuring the edge support deep learning model based on the model information, support method. 7. The method of claim 6,
acquiring, from the edge device, at least one of FLOPS (FLoating point Operations Per Second), RAM size, GPU size, image resolution, and processing speed, as performance information about the edge device; and
Further comprising the step of configuring the edge support deep learning model based on the performance information, support method. A computer-readable recording medium in which a program for performing the method according to claim 6 is recorded. A computer program stored in a medium for performing the support method according to claim 6, which is performed by a support device that communicates with an edge device that performs deep learning. A deep learning system comprising an edge device that performs deep learning, and a support device that communicates with the edge device, as a method for supporting deep learning of the edge device,
obtaining, by the support device, learning data from the edge device;
providing, by the support device, an edge support deep learning model learned based on the acquired learning data to the edge device; and
Supporting method, comprising the step of configuring, by the edge device, an edge-only deep learning model based on the edge supported deep learning model.

Images