KR20210076641A - Method for providing artificial intelligence service - Google Patents

Abstract

According to an embodiment of the present disclosure, a computer program stored in a computer-readable storage medium is disclosed. When the computer program is executed on one or more processors, the present invention performs the following operations for providing artificial intelligence services, the operations includes the steps of: transmitting a model to at least one edge device so that the edge device learns the model; receiving information about a learning result of the model from at least one edge device; and updating the model based on the learning result information received from at least one edge device.

Description

METHOD FOR PROVIDING ARTIFICIAL INTELLIGENCE SERVICE

The present disclosure relates to an information processing method using a computing device, and more particularly, to a technology for providing an artificial intelligence service.

In the rapidly developing IoT market, the current cloud computing structure may have limitations to meet the rapidly increasing traffic volume of edge devices and the increasing data demands of users. In the current cloud computing structure, various problems such as concentration of load, limitation of real-time connection, security problems, and complex database design may exist.

Therefore, in recent years, an edge computing structure to overcome these problems has been proposed. The edge computing structure may be a structure that performs a calculation process in a physically close gateway or computing device and provides a quick response in real time.

Currently, artificial intelligence service platforms are sometimes implemented through cloud computing structures. However, as described above, the cloud computing structure has a problem. Therefore, in recent years, attempts to borrow an edge computing structure for a stable artificial intelligence service platform are increasing.

Therefore, there may be a demand in the industry for a technology that provides a stable artificial intelligence service.

Korean Patent Publication No. 10-2019-0085823 discloses a personalized question-and-answer system capable of protecting personal information, a cloud server, and a method of providing a common neural network model thereof.

The present disclosure has been devised in response to the above-described background technology, and is intended to provide a method for providing an artificial intelligence service.

According to an embodiment of the present disclosure for realizing the above-described problems, a computer program including instructions stored in a computer-readable storage medium to cause the computer to perform the following operations is disclosed. The operations may include transmitting a model to at least one edge device so that the edge device learns the model; receiving information about a learning result of the model from at least one edge device; and updating the model based on the learning result information received from at least one edge device.

In an alternative embodiment, the operation of sending a model to at least one edge device to allow the edge device to learn the model may include transmitting a model requested by a user to the edge device based on a model request signal received from the edge device. It can include actions.

In an alternative embodiment, the learning result information of the model may include at least one of version information of the learned model, a weight of the learned model, an amount of weight change according to learning, or data related to a learning process of the model.

In an alternative embodiment, the weight of the trained model may reflect pattern information extracted from the training data set used to train the model on the edge device.

In an alternative embodiment, the operation of updating the model based on the learning result information received from the at least one edge device may include updating the model based on at least part of the weights of the learned model received from the at least one edge device. action; updating the model based on at least a portion of two or more learned models received from two or more edge devices; or updating version information of the model; may include at least one of

In an alternative embodiment, the method further includes performing a synchronization operation based on the state information, wherein the synchronization operation is to update the model in order to perform a service in which the server or at least one edge device provides the same model to the user. may include tasks.

In an alternative embodiment, the synchronization operation is a first synchronization operation to send the model to the failed edge device, a second synchronization operation to cause another edge device to send the model to the failed edge device, or the model from the edge device. It may include at least one of the third synchronization task for receiving.

In an alternative embodiment, the state information may include at least one of version information of the model, meta information of learning data of the model, storage location information of the model, operation-related information of an edge device, or operation-related information of a server.

In an alternative embodiment, the first synchronization operation may include an operation of performing synchronization by transmitting a model to the edge device having a failure based on state information received from a coordinator module mapped with the edge device having a failure.

In an alternative embodiment, the first synchronization operation is performed, when the version of the model stored in the faulty edge device is lower than the version of the model stored in the server, the model stored in the server is mapped to the faulty edge device by the coordinator module or the faulty sending to at least one of the generated edge devices to cause the faulty edge device to update the model stored in the server; Alternatively, if the version of the model stored in the faulty edge device is lower than the version of the model stored in the other edge device, the other edge device can convert the model stored in the other edge device to the coordinator module mapped with the faulty edge device or the faulty edge. sending to at least one of the devices to cause the failed edge device to update the model stored in the other edge device; may include at least one of

In an alternative embodiment, the second synchronization operation includes performing synchronization by causing another edge device to transmit a model to the failed edge device based on the status information received from the coordinator module mapped with the failed server. can do.

In an alternative embodiment, in the second synchronization operation, when the version of the model stored in the faulty edge device is lower than the version of the model stored in the other edge device, the model stored in the other edge device is mapped to the faulty edge device. It may include the operation of sending to at least one of the coordinator module or the faulty edge device to cause the faulty edge device to update the model stored in the other edge device.

In an alternative embodiment, the third synchronization operation may include an operation of receiving a model from an edge device or a coordinator module mapped with an edge device based on state information received from a coordinator module mapped with a server in which a failure occurred.

In an alternative embodiment, in the third synchronization operation, when the version of the model stored in the edge device is higher than the version of the model stored in the server with the failure, the model stored in the edge device is mapped to the edge device or the edge device and the coordinator module The operation of receiving from at least one of the following and updating the model stored in the edge device may be further included.

In an alternative embodiment, the edge device includes: a weighted edge device capable of performing training of the model and transmission of the model; or a lightweight edge device capable of performing model transmission; may include at least one of

In an alternative embodiment, the transmitting of the model may include transmitting to the user terminal via at least one of the at least one said weight edge device or at least one said lightweight edge device.

According to another embodiment of the present disclosure, a method for doing so is disclosed. transmitting the model to at least one edge device to cause the edge device to train the model; Receiving learning result information of the model from at least one edge device; and updating the model based on the learning result information received from at least one edge device.

A server for performing according to another embodiment of the present disclosure is disclosed. one or more processors; and a memory storing instructions executable by the one or more processors; including, wherein the one or more processors transmit a model to at least one edge device to cause the edge device to learn the model, receive information about a learning result of the model from at least one edge device, and at least The model may be updated based on the learning result information received from one edge device.

According to an embodiment of the present disclosure, a server for providing an artificial intelligence service may be provided.

BRIEF DESCRIPTION OF THE DRAWINGS So that the above-mentioned features of the present disclosure may be understood in detail, with a more specific description, with reference to the following embodiments, some of the embodiments are shown in the accompanying drawings. Also, like reference numerals in the drawings are intended to refer to the same or similar functions throughout the various aspects. However, it should be noted that the accompanying drawings only show certain typical embodiments of the present disclosure and are not to be considered as limiting the scope of the present disclosure, and other embodiments having the same effect may be fully appreciated. Take note.
1 is a block diagram of a server for providing an artificial intelligence service according to an embodiment of the present disclosure.
2 is an exemplary diagram schematically illustrating a network structure of a model provided in the process of providing an artificial intelligence service.
3 is an exemplary diagram for explaining a method for providing an artificial intelligence service.
4 is an exemplary diagram for explaining a synchronization operation performed when a failure occurs in an edge device in the process of providing an artificial intelligence service.
5 is an exemplary diagram for explaining a synchronization operation performed when a failure occurs in a server in the process of providing an artificial intelligence service.
6 is an exemplary diagram illustrating a multiplexing network for providing an uninterrupted artificial intelligence service.
7 is a flowchart for providing an artificial intelligence service.
8 is a block diagram illustrating a module for providing an artificial intelligence service.
9 is a simplified, general schematic diagram of an exemplary computing environment in which embodiments of the present disclosure may be implemented.

Various embodiments are now described with reference to the drawings. In this specification, various descriptions are presented to provide an understanding of the present disclosure. However, it is evident that these embodiments may be practiced without these specific descriptions.

The terms “component,” “module,” “system,” and the like, as used herein, refer to a computer-related entity, hardware, firmware, software, a combination of software and hardware, or execution of software. For example, a component can be, but is not limited to being, a process running on a processor, a processor, an object, a thread of execution, a program, and/or a computer. For example, both an application running on a server and a server can be components. One or more components may reside within a processor and/or thread of execution. A component may be localized within one computer. A component may be distributed between two or more computers. In addition, these components can execute from various computer readable media having various data structures stored therein. Components may communicate via a network such as the Internet with another system, for example via a signal having one or more data packets (eg, data and/or signals from one component interacting with another component in a local system, distributed system, etc.) may communicate via local and/or remote processes depending on the data being transmitted).

In addition, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or.” That is, unless otherwise specified or clear from context, "X employs A or B" is intended to mean one of the natural implicit substitutions. That is, X employs A; X employs B; or when X employs both A and B, "X employs A or B" may apply to either of these cases. It should also be understood that the term “and/or” as used herein refers to and includes all possible combinations of one or more of the listed related items.

Also, the terms "comprises" and/or "comprising" should be understood to mean that the feature and/or element in question is present. However, it should be understood that the terms "comprises" and/or "comprising" do not exclude the presence or addition of one or more other features, elements and/or groups thereof. Also, unless otherwise specified or unless the context is clear as to designating a singular form, the singular in the specification and claims should generally be construed to mean "one or more."

In addition, the term "at least one of A or B" should be interpreted as meaning "when including only A", "when including only B", and "when combined with the configuration of A and B".

Those skilled in the art will further appreciate that the various illustrative logical blocks, configurations, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be implemented in electronic hardware, computer software, or combinations of both. It should be recognized that they can be implemented with To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. However, such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

Descriptions of the presented embodiments are provided to enable those skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art of the present disclosure. The generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present invention is not limited to the embodiments presented herein. This invention is to be construed in its widest scope consistent with the principles and novel features presented herein.

1 is a block diagram of a server for providing an artificial intelligence service according to an embodiment of the present disclosure.

The configuration of the server 100 shown in FIG. 1 is only a simplified example. In an embodiment of the present disclosure, the server 100 may include other components for performing the computing environment of the server 100 , and only some of the disclosed components may configure the server 100 .

According to an embodiment of the present disclosure, the server 100 may include a processor 110 , a memory 130 , and a network unit 150 .

According to an embodiment of the present disclosure, an edge device has a mechanism for communicating with each other or other nodes through a communication network, and may mean any type of node in a system for storing and utilizing data. For example, an edge device may include a PC, laptop computer, workstation, terminal, and/or any electronic device with network connectivity. Also, the edge device may include an arbitrary server implemented by at least one of an agent, an application programming interface (API), and a plug-in. Additionally, an edge device may include an application source and/or a client application. The above-described edge device is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, an edge device may be an entry point that provides access to a core network of an enterprise or a service provider. The edge device may include, for example, a router, a routing switch, an integrated access device (IAD), a multiplexer, a metropolitan area network (MAN), a wide area network (WAN) access device, and the like. The above-described edge device is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the artificial intelligence service may include a service for transmitting the model to the user terminal when the processor 110 receives a model transmission request signal from the user terminal. In addition, when the processor 110 receives a model transmission request signal from the user terminal, the artificial intelligence service may cause the edge device to transmit the model to the corresponding user terminal. Accordingly, the user may be provided with a necessary model by the processor 110 providing the artificial intelligence service. The aforementioned artificial intelligence service is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 110 may provide an artificial intelligence service. The processor 110 may receive a model transmission request from the edge device through the network unit 150 in order to provide an artificial intelligence service. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 110 may cause the edge device to train the model by transmitting the model to at least one edge device in order to provide an artificial intelligence service. The processor 110 may transmit the model requested by the user to the edge device based on the model request signal received from the edge device. A model may be a network function that performs inference by extracting regularities from a data set. The model may include a deep learning model that performs inference by extracting regularities from a data set. The model may also include a machine learning model. The processor 110 may obtain an inference value after extracting the regularity existing in the data set using the model. For example, the processor 110 may extract scalability existing in a plurality of vehicle images using a model. The processor 110 may obtain an inference value (eg, the probability that the object included in the image is a car 80&) as to whether the image input to the model is a car image based on the regularity obtained using the model. can When the network unit 150 receives the model transmission request, the processor 110 may transmit the requested model to the corresponding edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 110 may receive model learning result information from at least one edge device. The learning result information of the model may include information related to the result of learning the model. The learning result information of the model may include information generated as a result of learning the model and/or information generated as a result of learning the model. The learning result information of the model may include at least one of version information of the learned model, weights of the learned model, an amount of weight change according to learning, or data related to a learning process of the model. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the version information of the learned model may include information for identifying when the learned model is updated. The version information of the learned model may include information on an updated date, time, an update subject, and an updated location (eg, a server or an edge device). Therefore, the version information of the model may include "version name v2, updated date: 2019-10-12, update subject: administrator, updated location: server". As another example, the version information of the model may include "model name: CNN (Convolutional Neural Network), version name v4, updated date: 2019-12-12, updated subject: user, updated location: edge device". One item is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the weight of the learned model may include the weight of the model that has been trained. In the training process of the model, the weights of the model may be updated. When the training of the model is completed, the weight of the model may not change any more. The processor 110 may receive the weight of the learned model from the edge device. According to an embodiment of the present disclosure, the weight of the learned model may reflect pattern information extracted from the training data set used to train the model in the edge device. The pattern information may include information about data that appears repeatedly with a certain regularity in a data set. The pattern information may include, for example, information on a correlation between at least one or more attributes. Also, the pattern information may include, for example, information about a criterion for classifying data. The processor 110 may predict a continuous or discontinuous variable using a model including pattern information. Also, the processor 110 may cluster data having similar properties by using a model including pattern information. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the weight of the learned model may reflect pattern information extracted from the training data set used to train the model in the edge device. The processor 110 may generate an inference value by using pattern information reflected in the weight of the model. For example, the training data set may include image data of a person's face taken from various angles. In this case, the pattern information reflected in the weight of the model may be pattern information on the face shape of a specific person. Accordingly, the processor 110 may obtain an inference value for confirming whether a face of a specific person is a face of a specific person with respect to the newly input human face image by using the weight of the learned model that reflects the face pattern information of the specific person. Therefore, the model trained using the training data set on a specific edge device can receive and use the model in which the learning result (pattern information) is reflected in the weight of the trained model even if the server and/or other edge device do not use the corresponding training data set. can Therefore, it may be possible to share a learned model with knowledge of the corresponding training data set without transmitting and receiving the training data set. Therefore, it is possible to share the learned model reflecting the knowledge of the training data set that requires high security without sending and receiving training data sets that require high security (eg, medical image data, data related to state secrets, etc.). Therefore, high security can be maintained in the process of providing artificial intelligence services. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, a weight change amount according to learning may mean a weight change amount changed for each learning step (eg, a learning epoch). For example, the difference between the model weight of epoch 1 and the model weight of epoch 2 may be a weight change amount according to learning. The weight change amount according to learning may mean a weight change amount that is changed according to a plurality of learning steps. For example, the weight change according to learning may be a difference between the weight of a model that has undergone a learning process of 5 epochs and a weight of a model that has undergone a learning process of 10 epochs. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the data related to the learning process of the model may include data related to the learning process of the model. The data related to the learning process of the model may be data related to the hyperparameter of the model. Data related to the learning process of the model may include a learning epoch, a learning rate for each epoch, a learning time required, a bias, and the like. By receiving data related to the learning process of the model, the processor 110 may acquire information on how the model was trained. The processor 110 may retrain the model using data related to the learning process of the model. In addition, the processor 110 may transmit data related to the learning process of the model to another edge device. Through this, other users can acquire necessary information in the process of training the model. By sharing data related to the training process of the model, the time required for hyperparameter tuning of the model can be shortened. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to the present disclosure, the processor 110 may receive the learning result information of the model from at least one edge device. The processor 110 may not receive the training data set used for training the model while receiving the learning result information of the model. The training data set may be a data set in which security is very important, such as data related to a national security field or medical image information data. When these security-critical data sets are shared without restrictions, various problems such as invasion of privacy and exposure of national secrets may occur. Therefore, according to the present disclosure, it may be very important in terms of data security for the processor 110 to receive only the learning result information without receiving the training data set.

Hereinafter, an operation of updating the model based on the learning result information received from the edge device will be described in detail.

According to an embodiment of the present disclosure, the processor 110 may update the model based on the learning result information received from at least one edge device. The learning result information may include information related to a result of learning the model. The processor 110 may acquire a model from which various knowledge has been learned by updating the model based on the learning result information received from the edge device. That is, by receiving and updating the model learned from different edge devices, it is possible to acquire a model that has learned a wide range of knowledge without investing time and money in collecting learning data. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the model may be updated based on at least a part of weights of the learned model received from at least one edge device. The processor 110 may transmit the model A to the edge device 1 . In addition, the processor 110 may receive the model A' learned from the edge device 1 . The processor 110 may update a part of the weight of the model A by using the model A'. The processor 110 may perform an update in which at least some of the weights of the model A are changed to the weights of the model A'. The processor 110 may update the model A by replacing the model A with the model A'. The processor 110 may evaluate the performance of the model A' using the test data set. When the performance of the model A' obtained by inputting the test data set is superior to that of the model A, the processor 110 may perform an update in which the model A is replaced with the model A'. If the performance of the model A' is lower than that of the model A, the processor 110 may not perform the update on the model A. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 110 may update the model based on the learning result information received from at least one edge device. The processor 110 may update the model based on at least a portion of the two or more learned models received from the two or more edge devices. The processor 110 may update the model stored in the server by using only a partial model of each learned model received from each edge device. The processor 110 may determine the importance for each learned model received from each edge device, and determine whether to reflect the learned model to the model stored in the server based on the importance. In addition, the processor 110 may determine how much the learning degree of each model is reflected in the model stored in the server according to the importance of each learned model received from each edge device. According to an embodiment of the present disclosure, the model transmitted by the processor 110 to the edge device 1, the edge device 2, and the edge device 3 may be the same as the model A. In addition, the model learned from the edge device 1 may be model A1, the model learned from the edge device 2 may be model A2, and the model learned from the edge device 3 may be model A3. The processor 110 may update the model A to the model A4 by merging the weight of the learned model A1, the weight of the learned model A2, and the weight of the learned model A3. In this case, the weight of the learned model A1, the weight of the learned model A2, and the weight of the learned model A3 may be updated by giving importance to each weight. Accordingly, the processor 110 may obtain a model A4 in which different importance is assigned to each of the weight of the learned model A1, the weight of the learned model A2, and the weight of the learned model A3. The processor 110 may test the performance of the model A4 using the test data set. For example, the processor 110 may give high importance to the weight of the learned model A1 and give low importance to the weight of the remaining learned model A2 and the weight of the learned model A3 . In this case, the performance of the Model A4 may be low. However, the processor 110 may give high importance to the weight of the model A2 and give low importance to the remaining weights of the learned model A1 and the weights of the learned model A3 . In this case, the performance of the Model A4 may be high. Accordingly, the processor 110 may update the model A by finally giving a high importance to the weight of the model A2, and giving a low importance to the weight of the remaining learned model A1 and the weight of the learned model A3. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to another embodiment of the present disclosure, the processor 110 may update the model based on at least a portion of the two or more learned models received from the two or more edge devices. According to an embodiment of the present disclosure, the models transmitted by the processor 110 to the edge device 1, the edge device 2, and the edge device 3 may be different as Model A, Model B, and Model C, respectively. The model trained on the edge device 1 may be model A1, the model trained on the edge device 2 may be model B2, and the model trained on the edge device 3 may be model C3. The processor 110 may generate a model D in which the model learned from the edge device 1 is model A1, the model learned from the edge device 2 is model B2, and the model learned from the edge device 3 is model C3. Model D may be an ensemble model that finds an optimal answer (inference value) using multiple models. Accordingly, in the dog and cat classification problem, the processor 110 may determine that the inference value obtained using the models A1 and B2 is “dog” and the inference value obtained using the model C3 is “cat”. In this case, the processor 110 may determine that the inference value obtained using the model D is "dog" according to the principle of majority vote. The foregoing is merely an example, and the present disclosure is not limited thereto. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 110 may update the model based on the learning result information received from at least one edge device. The processor 110 may update version information of the model. The version information of the model may include information for identifying when the learned model is updated. The version information of the learned model may include information on an updated date, time, an update subject, an updated location (eg, a server or an edge device), and the like. When the processor 110 updates the model based on the learning result received from the edge device, version information of the model may also be updated. This is because it is necessary to update the version information of the model to determine whether the model stored in the server and/or edge device is the most updated model in common. That is, even for the same model A, the weight of the model A1 learned from the edge device 1 and the weight of the model A2 learned from the edge device 2 may be different. Through this, the processor 110 may provide the user with the latest updated model in common. The foregoing is merely an example, and the present disclosure is not limited thereto.

Hereinafter, an operation of performing a synchronization operation based on state information will be described in detail.

According to an embodiment of the present disclosure, the processor 110 may perform a synchronization operation based on state information. The synchronization operation may include maintaining the model stored in the server and at least one edge device as the most recently updated model. For example, when the model A stored in the server is the model A1 updated on 10/10/2019, the synchronization operation may include a task of allowing the model stored in at least one edge device to be A1. Through this, both the server and at least one edge device may use the same model or provide it to different edge devices. Through this, even when a problem occurs in the server and/or the edge device, the processor 110 may continuously provide the model to the user without interruption of the AI service. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the synchronization operation may include a first synchronization operation to transmit a model to a faulty edge device, a second synchronization operation to cause another edge device to transmit a model to a faulty edge device, or an edge and at least one of a third synchronization operation of receiving the model from the device. According to an embodiment of the present disclosure, a faulty edge device may include a terminal that cannot perform a service providing the most recently updated model. For example, the server may have model A1 updated on 10/10/2019 for model A. On the other hand, the faulty edge device may store the model A2 updated on 2016-10-10 with respect to the same model A. That is, the edge device 310 in which a failure has occurred may not store the most recently updated model. Accordingly, the faulty edge device may not be able to perform the service of providing the most recently updated model A1 to the user. According to another embodiment of the present disclosure, the faulty edge device may be in a state in which communication with the server and other edge devices is impossible due to a problem in the computing device. Also, the faulty edge device may be an edge device that cannot perform model update due to a problem in the computing device. Also, the faulty edge device may be an edge device that has lost the latest version model stored in the memory due to a problem in the computing device. Accordingly, the processor 110 transmits the model to the faulty edge device or causes another edge device to transmit the model to the faulty edge device, so that the faulty edge device also provides the most recently updated model. can be performed. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 110 may perform a synchronization operation based on state information. The status information may include status information of the server and/or edge device. For example, the status information may include information on whether the server and/or the edge device is operating normally. The state information may also include state information of the model stored by the server and/or the edge device. Accordingly, the model state information may include information on whether the model stored by the server and/or the edge device is the most recently updated model. The processor 110 may determine whether the model stored in the memory is the most recently updated model based on the state information of the model. Specifically, the processor 110 may determine whether the model stored in the edge device is the most recently updated model based on the state information of the edge device. Also, the processor 110 may determine whether a model stored in a specific edge device is the same as a model stored in another edge device based on the state information of the edge device.

According to an embodiment of the present disclosure, the state information may include at least one of version information of the model, meta information of learning data of the model, storage location information of the model, operation-related information of an edge device, or operation-related information of a server . According to an embodiment of the present disclosure, the version information of the model may include information for identifying when the learned model is updated. The version information of the learned model may include information on an updated date, time, an update subject, and an updated location (eg, a server or an edge device). Therefore, the version information of the model may include "version name v2, updated date: 2019-10-12, update subject: administrator, updated location: server". According to an embodiment of the present disclosure, the learning data meta information of the model may include knowledge information possessed by the learning data. For example, the training data meta information of the model may include information that is a training data set for classifying dogs and cats. In addition, the learning data meta-information of the model includes the purpose to be achieved using the model (for example, the purpose of classifying dogs and cats, the purpose of performing face recognition, the purpose of performing intention analysis included in the sentence) can do. According to an embodiment of the present disclosure, the learning data meta information of the model may include information on the form of the training data (eg, whether image data, voice data, time series data, or text data). According to an embodiment of the present disclosure, the storage location information of the model may include information on which computing device the model is located. The model storage location information may include information on which server and/or which edge device the model is stored. In addition, the storage location information of the model may include information (eg, an index address value, an address value in a physical space) located in a memory space of a specific server and/or a specific edge device. According to an embodiment of the present disclosure, the operation-related information of the edge device and/or the operation-related information of the server may be information necessary to detect whether a failure occurs in the edge device and/or the server. The operation-related information of the edge device and/or the server may include, for example, CPU computation amount, memory usage, communication status, number of packets exchanged per hour, and the like. The foregoing is merely an example, and the present disclosure is not limited thereto.

Hereinafter, a description of a process of transmitting and receiving state information in order to perform a synchronization operation will be described in detail with reference to FIG. 3 .

In FIG. 3 , the server 100, the edge devices 210, 230, and 250, and the coordinator module mapped to each server 100 and/or the edge device 210, 230, 250, learning result information 270, state Information 290 is shown.

According to an embodiment of the present disclosure, the coordinator module may include a module physically independent from the server 100 or the edge devices 210 , 230 , and 250 . Accordingly, the coordinator module can operate normally even when a failure occurs in the server 100 or the edge devices 210 , 230 , and 250 . The coordinator module may receive status information of the mapped server or edge device and transmit the information to a physically independent device such as another server, another edge device, or another coordinator module. There may be only one coordinator module, or a plurality of coordinator modules may exist. When only one coordinator module exists, the corresponding coordinator module may have a one-to-many mapping relationship with the server and/or all edge devices. When a plurality of coordinator modules exist, each of the server and/or edge device may have a one-to-one mapping relationship. The above-described mapping relationship is merely an example, and the present disclosure is not limited thereto. According to another embodiment of the present disclosure, the coordinator module may be a module included in the server and/or the edge device. The above-described coordinator module is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the processor 110 may receive the state information 290 from the server 100 and the mapped coordinator module. The corresponding state information 290 may be state information of the edge device 210 . The processor 110 may receive state information of the edge devices 210 , 230 , and 250 from a coordinator module mapped with the edge devices 210 , 230 , and 250 , respectively. In addition, the processor 110 may receive the state information of the edge devices 210 , 230 , and 250 from the coordinator module mapped with the server 100 . The coordinator module may acquire status information of the server and/or edge device through monitoring in real time. According to an embodiment of the present disclosure, the processor 110 may receive the learning result information 270 from the edge device 210 and/or the coordinator module mapped with the edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

Hereinafter, a synchronization operation performed when a failure occurs in the edge device will be described in detail with reference to FIG. 4 .

According to an embodiment of the present disclosure, the processor 110 may perform a first synchronization operation of transmitting a model to the edge device 310 having a failure. The first synchronization operation may include an operation of performing synchronization by transmitting (313) a model to the edge device having a failure based on state information received from the coordinator module mapped with the edge device 310 in which the failure occurred. The processor 110 may receive the abnormal state information 311 from the faulty edge device 310 and/or the coordinator module mapped with the faulty edge device 310 . In this case, the processor 110 may directly receive the abnormal state information 311 , or may receive the abnormal state information 311 indirectly (bypass) through the coordinator module mapped with the server 100 . The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, there may exist a case in which the version of the model stored in the faulty edge device 310 is lower than the version of the model stored in the server. The version information of the model may include information for identifying when the learned model is updated. The processor 110 may compare the version of the model based on the version information of the model. For example, for the same model A, if model A was updated on 2015-01-01, model version information is model A-v1, updated: 2015-01-01, model A updated on 2019-01-01 If it is, the model version information may be model A-v3, updated date: 2019-01-01. Accordingly, the processor 110 may compare the versions of the model A-v1 and the model A-v3 to determine that the model A-v3 is a higher version compared to the model A-v1. Cases in which the edge device fails may include cases in which the latest model cannot be transmitted from the server 100, cases in which the edge device itself cannot update the model, and the like. In this case, there may be a need to update the model to the latest model for an AI service that can continuously provide the latest model. Therefore, the processor 110 transmits the model stored in the server 100 to at least one of the faulty edge device 310 and the mapped coordinator module or the faulty edge device 310 to cause the faulty edge device 310 to the server. may cause it to update to the model stored in . Here, an error may occur in the network unit of the edge device 310 having a failure, and direct communication with the server may not be possible. In this case, the server may transmit the model to the faulty edge device 310 through the coordinator module mapped with the faulty edge device 310 . The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, there may exist a case where the version of the model stored in the faulty edge device 310 is lower than the version of the model stored in other edge devices. The processor 110 may compare the version of the model stored in each edge device based on the version information of the model. Accordingly, the processor 110 causes the other edge devices 230 and 250 to transfer the model stored in the other edge devices 230 and 250 to the faulty edge device 310 and the mapped coordinator module or the faulty edge device 310 . It can be transmitted to at least one to cause the faulty edge device 310 to update the model stored in the other edge device. Here, an error may occur in the network unit of the edge device 310 having a failure, and direct communication with other edge devices 230 and 250 may not be possible. In this case, the server 100 may cause other edge devices 230 and 250 to transmit the model to the edge device 310 in which the failure occurs through the coordinator module mapped with the edge device 310 in which the failure occurred. Accordingly, the server 100 may cause the faulty edge device to update the model stored in the other edge devices 230 and 250 . Through this, a sudden increase in the load on the server is reduced, so that the speed at which the server transmits the model and the speed at which the server processes the task is maintained constant without abruptly decreasing. The foregoing is merely an example, and the present disclosure is not limited thereto.

Hereinafter, a synchronization operation performed when a failure occurs in the server will be specifically described with reference to FIG.

According to an embodiment of the present disclosure, the synchronization operation may include a second synchronization operation that causes the other edge devices 230 and 250 to transmit a model to the edge device 310 where the failure occurs. The second synchronization operation causes the other edge devices 230 and 250 to transmit a model to the edge device 310 where the failure occurs based on the status information received from the coordinator module mapped with the failed server, thereby performing synchronization. may include The server 100 having a failure may include a server that cannot perform a service providing the most recently learned model. For example, the edge devices 230 and 250 may have model A1 updated on 10/10/2019 with respect to model A. On the other hand, the server 100 having a failure may store the model A2 updated on 10/10/2016 for the same model A. That is, the server 100 having a failure may not store the most recently updated model. Accordingly, the server 100 having a failure may not be able to perform a service of providing the most recently learned model A1 to other edge devices. According to another embodiment of the present disclosure, the server 100 having a failure may be in a state in which communication with other servers and edge devices is impossible due to a problem in the computing device. Also, the server 100 having a failure may be a server that cannot perform model update due to a problem in the computing device. Also, the faulty edge device may be an edge device that has lost the latest version model stored in the memory due to a problem in the computing device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the second synchronization operation causes other edge devices 230 and 250 to model the faulty edge device based on the status information received from the coordinator module mapped with the faulty server 100. may include the task of performing synchronization by causing the In this case, since a failure has occurred in the server, the processor 110 may not be able to directly transmit the model to the edge device. As another example, since a failure occurs in the server, there may exist a case where the version of the model stored in the faulty edge device 310 is lower than the version of the model stored in the other edge devices 230 and 250 . The processor 110 may compare the version of the model stored in each edge device based on the version information of the model. In this case, the processor 110 transmits the model stored in the other edge devices 230 and 250 to at least one of the faulty edge device 310 and the mapped coordinator module or the faulty edge device 310 to the faulty edge device. may cause 310 to update with the model stored in other edge devices 230 , 250 . Through this, artificial intelligence services can be provided without interruption even when the server fails. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the third synchronization operation may include an operation of receiving a model from an edge device or a coordinator module mapped with an edge device based on state information received from a coordinator module mapped with a server with a failure. have. By performing the third synchronization operation, the server 100 having a failure may update the model to the latest version of the model. Furthermore, the server 100 in which the failure has occurred may store the latest version of the model in the edge device. Through this, it is possible to provide a continuous artificial intelligence service by normalizing the server 100 in which the failure occurs in a short time. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, in the third synchronization operation, when the version of the model stored in the edge device 230 is higher than the version of the model stored in the server 100 in which the failure occurs, the model stored in the edge device 230 is higher. The method may further include the operation of receiving from at least one of the edge device 230 or the coordinator module mapped with the edge device 230 and updating the model with the model stored in the edge device 230 . There may exist a case where the version of the model stored in the edge device 230 is higher than the version of the model stored in the server 100 in which the failure occurs. The processor 110 may compare the version of the model stored in each edge device based on the version information of the model. When a failure occurs in the server 100 , the processor 110 may not be able to update the model. Therefore, the version of the model stored in the server 100 in which the failure occurs may be lower than the version of the model stored in the edge device. In this case, the processor 110 may receive 450 the model stored in the edge device 230 from at least one of the edge device 230 or a coordinator module mapped with the edge device 230 . At this time, the processor 110 may receive the model stored in the edge device 230 through the coordinator module mapped with the server 100 in which the failure occurred. The reason is that there may exist a case where the server 100 in which the failure occurs cannot directly receive the model from the edge device 230 . The processor 110 may receive the model stored in the edge device 230 and update the model. After updating the model, the processor 110 may transmit the updated model to the edge device 310 in which a failure occurs. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the edge device may include at least one of a weight edge device and a lightweight edge device. The weighted edge device may train the model, and may transmit the model to a server and/or other edge device. The processor 110 may receive the model request signal from the user terminal to cause the weight edge device to transmit the model to the corresponding user terminal. Lightweight edge devices cannot train models, but can send models to servers and/or other edge devices. The processor 110 may receive the model request signal from the user terminal to cause the lightweight edge device to transmit the model to the corresponding user terminal. There may be a coordinator module mapped to the lightweight edge device and/or the heavy edge device, respectively. According to an embodiment of the present disclosure, the lightweight edge device may transmit/receive a model to and from the weight edge device. In addition, the weight edge device can send and receive models to and from the server. According to another embodiment of the present disclosure, the lightweight edge device may transmit/receive a model directly to/from the server. Model transmission/reception may be performed through a coordinator module mapped to each of the server, the heavy edge device, and the lightweight edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

Hereinafter, a description of a multiplexed network for providing an uninterrupted artificial intelligence service will be described in detail with reference to FIG. 6 .

According to an embodiment of the present disclosure, the server 100 may be connected to at least one edge device. Each of the at least one edge device may be connected to the at least one edge device. Through this, a multiplexed network more than triplex can be implemented in the artificial intelligence service providing system. In a multiplexed connection network, a server and/or an edge device may be expressed as a vertex, and connection information of each vertex may be expressed as a trunk line. When an edge exists between the vertex and the vertex, the connected vertex may transmit/receive information about the model and/or the learning result of the model. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, transmitting the model may include transmitting the model to the user terminal via at least one of at least one heavy edge device and at least one lightweight edge device.

According to an embodiment of the present disclosure, the server 100 may cause the lightweight edge device to transmit the model to the faulty lightweight edge device and/or the faulty heavy edge device. In addition, the server 100 may cause the heavy edge device to transmit the model to the faulty weight edge device and/or the faulty lightweight edge device. The failed server may receive the model from the lightweight edge device and/or the heavy edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, at least one heavy edge device may be present between the server and the lightweight edge device or may be omitted. As shown in FIG. 6 , a heavy edge device 1 720 may exist between the server 100 and the lightweight edge device 2 721 . According to another embodiment of the present disclosure, at least one lightweight edge device may exist between the server 100 and the lightweight edge device 4 751 or may be omitted. According to another embodiment of the present disclosure, at least one other lightweight edge device and at least one weight edge device may exist between the server 100 and the lightweight edge device 4 751 . According to another embodiment of the present disclosure, at least one other weight edge device 2 730 may exist or may be omitted between the server 100 and the weight edge device 3 731 . Also, at least one other edge device may exist between the server and the weight edge device or may be omitted. There may be at least one lightweight edge device and at least one weight edge device between the server and the weight edge device. Through this, multiplexed connection more than tripled can be implemented in the artificial intelligence service providing system. Accordingly, the processor 110 may provide the model to the user without interruption through distributed processing even when a failure occurs in the server and/or the edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the number of edge devices existing between the server and the edge device may be defined as a hop. In this case, edge devices present in hop 0 are lightweight edge device 1 710 , weight edge device 1 720 , weight edge device 2 730 , weight edge device 4 740 , and weight edge device 5 750 . can exist. Edge devices existing in hop 0 can transmit and receive models to each other. Accordingly, the processor 110 may cause the normal 0-hop edge device to transmit the model to the faulty edge device. The one-hop edge device may be a lightweight edge device 2 721 and a heavy edge device 3 731 . In a multiplexed network for providing artificial intelligence services, hops may be n-hops. In this case, n may be any positive integer. Therefore, the processor ( ) can guarantee high availability by providing artificial intelligence services in an n-hop multiplexed network. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the first synchronization operation includes performing synchronization by transmitting a model to the lightweight edge device in which the failure occurs based on the state information received from the coordinator module mapped to the lightweight edge device in which the failure occurs. can do. The server may cause the lightweight edge device and/or the heavy edge device to send the model to the failed lightweight edge device. The server can also send the model directly to the failed lightweight edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the first synchronization operation includes the operation of performing synchronization by transmitting a model to the weight edge device in which the failure occurs based on the state information received from the coordinator module mapped with the weight edge device in which the failure occurs. can do. The server may cause the lightweight edge device and/or the heavy edge device to send the model to the failed heavyweight edge device. The server can also send the model directly to the failed heavy edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the second synchronization operation causes another heavy edge device and/or a lightweight edge device to cause a failure based on the state information received from the coordinator module mapped with the failed server. This may include causing the lightweight edge device to send the model to perform synchronization. A failed server may include a server that cannot perform a service providing the most recently trained model. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the third synchronization operation is performed on the basis of the state information received from the failed server and the mapped coordinator module, the weight edge device, the lightweight edge device, the weight edge device and the mapped coordinator module and/or the lightweight It may include the operation of receiving the model from the edge device and the mapped coordinator module. By performing the third synchronization operation, the failed server can update the model to the latest version of the model. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to the present disclosure, it is possible to provide an uninterrupted artificial intelligence service to a user by providing an artificial intelligence service through a synchronization operation. That is, in the event of a server failure, the AI service may be temporarily suspended until the server is restored. However, according to the present disclosure, an uninterrupted artificial intelligence service may be realized by receiving a model from another edge device and updating the model by receiving a model from a normal edge device and a server having a failure. In addition, since model learning is performed not only on the server but also on the edge device, the amount of computation and memory usage of the server can be reduced. And since a specific edge device may provide a model to other edge devices, the load on the server may be reduced. This is because, if the edge device cannot provide a model to other edge devices, the server load increases, which may cause a slowdown in the provision of artificial intelligence services.

2 is an exemplary diagram schematically illustrating a network structure of a model provided in the process of providing an artificial intelligence service.

Throughout this specification, computational model, neural network, network function, and neural network may be used interchangeably. A neural network may be composed of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. A neural network is configured by including at least one or more nodes. Nodes (or neurons) constituting the neural networks may be interconnected by one or more links.

In the neural network, one or more nodes connected through a link may relatively form a relationship between an input node and an output node. The concepts of an input node and an output node are relative, and any node in an output node relationship with respect to one node may be in an input node relationship in a relationship with another node, and vice versa. As described above, an input node-to-output node relationship may be created around a link. One or more output nodes may be connected to one input node through a link, and vice versa.

In the relationship between the input node and the output node connected through one link, the value of the output node may be determined based on data input to the input node. Here, a node interconnecting the input node and the output node may have a weight. The weight may be variable, and may be changed by the user or algorithm in order for the neural network to perform a desired function. For example, when one or more input nodes are interconnected to one output node by respective links, the output node sets values input to input nodes connected to the output node and links corresponding to the respective input nodes. An output node value may be determined based on the weight.

As described above, in a neural network, one or more nodes are interconnected through one or more links to form an input node and an output node relationship within the neural network. The characteristics of the neural network may be determined according to the number of nodes and links in the neural network, the correlation between the nodes and the links, and the value of a weight assigned to each of the links. For example, when the same number of nodes and links exist and there are two neural networks having different weight values between the links, the two neural networks may be recognized as different from each other.

A neural network may include one or more nodes. Some of the nodes constituting the neural network may configure one layer based on distances from the initial input node. For example, a set of nodes having a distance n from the initial input node may constitute n layers. The distance from the initial input node may be defined by the minimum number of links that must be passed to reach the corresponding node from the initial input node. However, the definition of such a layer is arbitrary for description, and the order of the layer in the neural network may be defined in a different way from the above. For example, a layer of nodes may be defined by a distance from the final output node.

The initial input node may mean one or more nodes to which data is directly input without going through a link in a relationship with other nodes among nodes in the neural network. Alternatively, in a relationship between nodes based on a link in a neural network, it may mean nodes that other input nodes connected by a link do not have. Similarly, the final output node may refer to one or more nodes that do not have an output node in relation to other nodes among nodes in the neural network. Also, the hidden node may refer to nodes constituting the neural network other than the first input node and the last output node. The neural network according to an embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be the same as the number of nodes in the output layer, and the number of nodes decreases and then increases again as the input layer progresses to the hidden layer. can In addition, the neural network according to another embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be less than the number of nodes in the output layer, and the number of nodes decreases as the number of nodes progresses from the input layer to the hidden layer. have. In addition, the neural network according to another embodiment of the present disclosure may be a neural network in which the number of nodes in the input layer may be greater than the number of nodes in the output layer, and the number of nodes increases as the number of nodes progresses from the input layer to the hidden layer. can The neural network according to another embodiment of the present disclosure may be a neural network in a combined form of the aforementioned neural networks.

A deep neural network (DNN) may refer to a neural network including a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks can be used to identify the latent structures of data. In other words, it can identify the potential structure of photos, texts, videos, voices, and music (e.g., what objects are in the photos, what the text and emotions are, what the texts and emotions are, etc.) . Deep neural networks include convolutional neural networks (CNNs), recurrent neural networks (RNNs), auto encoders, generative adversarial networks (GANs), and restricted Boltzmann machines (RBMs). boltzmann machine), a deep trust network (DBN), a Q network, a U network, a Siamese network, and the like. The description of the deep neural network described above is only an example, and the present disclosure is not limited thereto.

In an embodiment of the present disclosure, the network function may include an auto-encoder. The auto-encoder may be a kind of artificial neural network for outputting output data similar to input data. The auto encoder may include at least one hidden layer, and an odd number of hidden layers may be disposed between the input/output layers. The number of nodes in each layer may be reduced from the number of nodes in the input layer to an intermediate layer called the bottleneck layer (encoding), and then expanded symmetrically with reduction from the bottleneck layer to the output layer (symmetrical to the input layer). In this case, although the dimensionality reduction layer and the dimension reconstruction layer are illustrated as being symmetrical in the example of FIG. 2 , the present disclosure is not limited thereto, and the nodes of the dimension reduction layer and the dimension reconstruction layer may or may not be symmetrical. The auto-encoder can perform non-linear dimensionality reduction. The number of input layers and output layers may correspond to the number of sensors remaining after pre-processing of input data. In the auto-encoder structure, the number of nodes of the hidden layer included in the encoder may have a structure that decreases as the distance from the input layer increases. If the number of nodes in the bottleneck layer (the layer with the fewest nodes located between the encoder and decoder) is too small, a sufficient amount of information may not be conveyed, so a certain number or more (e.g., more than half of the input layer, etc.) ) may be maintained.

The neural network may be trained by at least one of supervised learning, unsupervised learning, and semi-supervised learning. The training of the neural network is to minimize the output error. In the training of a neural network, iteratively input the training data to the neural network, calculate the output and target errors of the neural network for the training data, and reduce the error of the neural network from the output layer of the neural network to the input layer. It is a process of updating the weights of each node of the neural network by backpropagation in the direction. In the case of teacher learning, learning data in which the correct answer is labeled in each learning data is used (ie, labeled learning data), and in the case of comparative learning, the correct answer may not be labeled in each learning data. That is, for example, learning data in the case of teacher learning related to data classification may be data in which categories are labeled in each of the learning data. Labeled training data is input to the neural network, and an error can be calculated by comparing the output (category) of the neural network with the label of the training data. As another example, in the case of comparison learning about data classification, an error may be calculated by comparing the input training data with the neural network output. The calculated error is back propagated in the reverse direction (ie, from the output layer to the input layer) in the neural network, and the connection weight of each node of each layer of the neural network may be updated according to the back propagation. A change amount of the connection weight of each node to be updated may be determined according to a learning rate. The computation of the neural network on the input data and the backpropagation of errors can constitute a learning cycle (epoch). The learning rate may be applied differently according to the number of repetitions of the learning cycle of the neural network. For example, a high learning rate can be used at the early stage of training of a neural network to increase efficiency by allowing the neural network to quickly acquire a certain level of performance, and a low learning rate can be used to increase accuracy at a later stage of learning.

In the training of neural networks, in general, the training data may be a subset of the actual data (that is, the data to be processed using the trained neural network), so the error on the training data is reduced but the error on the real data There may be increasing learning cycles. Overfitting is a phenomenon in which errors on actual data increase by over-learning on training data as described above. For example, a phenomenon in which a neural network that has learned a cat by showing a yellow cat does not recognize that it is a cat when seeing a cat other than yellow may be a type of overfitting. Overfitting can act as a cause of increasing errors in machine learning algorithms. In order to prevent such overfitting, various optimization methods can be used. In order to prevent overfitting, methods such as increasing training data, regularization, and dropout in which a part of nodes in the network are omitted in the process of learning may be applied.

A computer-readable medium storing a data structure is disclosed according to an embodiment of the present disclosure.

The data structure may refer to the organization, management, and storage of data that enables efficient access and modification of data. A data structure may refer to an organization of data to solve a specific problem (eg, data retrieval, data storage, and data modification in the shortest time). A data structure may be defined as a physical or logical relationship between data elements designed to support a particular data processing function. The logical relationship between data elements may include a connection relationship between data elements that a user thinks. Physical relationships between data elements may include actual relationships between data elements physically stored on a computer-readable storage medium (eg, a hard disk). A data structure may specifically include a set of data, relationships between data, and functions or instructions applicable to data. Through an effectively designed data structure, a computing device can perform an operation while using the resource of the computing device to a minimum. Specifically, the computing device may increase the efficiency of operations, reads, insertions, deletions, comparisons, exchanges, and retrievals through effectively designed data structures.

A data structure may be classified into a linear data structure and a non-linear data structure according to the type of the data structure. The linear data structure may be a structure in which only one piece of data is connected after one piece of data. The linear data structure may include a list, a stack, a queue, and a deck. A list may mean a set of data in which an order exists internally. The list may include a linked list. The linked list may be a data structure in which data is linked in such a way that each data is linked in a line with a pointer. In a linked list, a pointer may contain information about a link with the next or previous data. A linked list may be expressed as a single linked list, a doubly linked list, or a circularly linked list according to a shape. A stack can be a data enumeration structure with limited access to data. A stack can be a linear data structure in which data can be processed (eg, inserted or deleted) at only one end of the data structure. The data stored in the stack may be a data structure LIFO-Last in First Out. A queue is a data listing structure that allows limited access to data. Unlike a stack, the queue may be a data structure that comes out later (FIFO-First in First Out) as data stored later. A deck can be a data structure that can process data at either end of the data structure.

The nonlinear data structure may be a structure in which a plurality of data is connected after one data. The nonlinear data structure may include a graph data structure. A graph data structure may be defined as a vertex and an edge, and the edge may include a line connecting two different vertices. A graph data structure may include a tree data structure. The tree data structure may be a data structure in which one path connects two different vertices among a plurality of vertices included in the tree. That is, it may be a data structure that does not form a loop in the graph data structure.

Throughout this specification, computational model, neural network, network function, and neural network may be used interchangeably. (Hereinafter, the neural network is unified and described.) The data structure may include a neural network. And the data structure including the neural network may be stored in a computer-readable medium. Data structures, including neural networks, may also include data input to the neural network, weights of the neural network, hyperparameters of the neural network, data obtained from the neural network, activation functions associated with each node or layer of the neural network, and loss functions for learning the neural network. have. A data structure comprising a neural network may include any of the components disclosed above. That is, the data structure including the neural network includes all or all of the data input to the neural network, the weights of the neural network, the hyperparameters of the neural network, the data acquired from the neural network, the activation function associated with each node or layer of the neural network, and the loss function for training the neural network. may be configured including any combination of In addition to the above-described configurations, a data structure including a neural network may include any other information that determines a characteristic of a neural network. In addition, the data structure may include all types of data used or generated in the operation process of the neural network, and is not limited to the above. Computer-readable media may include computer-readable recording media and/or computer-readable transmission media. A neural network may be composed of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. A neural network is configured by including at least one or more nodes.

The data structure may include data input to the neural network. A data structure including data input to the neural network may be stored in a computer-readable medium. The data input to the neural network may include learning data input in a neural network learning process and/or input data input to the neural network in which learning is completed. Data input to the neural network may include pre-processing data and/or pre-processing target data. The preprocessing may include a data processing process for inputting data into the neural network. Accordingly, the data structure may include data to be pre-processed and data generated by pre-processing. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

The data structure may include data input to or output from the neural network. A data structure including data input to or output from the neural network may be stored in a computer-readable medium. The data structure stored in the computer-readable medium may include data input in an inference process of the neural network or output data output as a result of inference of the neural network. In addition, since the data structure may include data processed by a specific data processing method, it may include data before and after processing. Accordingly, the data structure may include data to be processed and data processed through a data processing method.

The data structure may include the weights of the neural network. (In this specification, a weight and a parameter may be used interchangeably.) And a data structure including a weight of a neural network may be stored in a computer-readable medium. The neural network may include a plurality of weights. The weight may be variable, and may be changed by the user or algorithm in order for the neural network to perform a desired function. For example, when one or more input nodes are interconnected to one output node by respective links, the output node sets values input to input nodes connected to the output node and links corresponding to the respective input nodes. An output node value may be determined based on the parameter. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

By way of example and not limitation, the weight may include a weight variable in a neural network learning process and/or a weight in which neural network learning is completed. The variable weight in the neural network learning process may include a weight at the start of the learning cycle and/or a variable weight during the learning cycle. The weight for which neural network learning is completed may include a weight for which a learning cycle is completed. Accordingly, the data structure including the weight of the neural network may include a data structure including the weight variable in the neural network learning process and/or the weight in which the neural network learning is completed. Therefore, it is assumed that the above-described weights and/or combinations of weights are included in the data structure including the weights of the neural network. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

The data structure including the weights of the neural network may be stored in a computer-readable storage medium (eg, memory, hard disk) after being serialized. Serialization can be the process of converting a data structure into a form that can be reconstructed and used later by storing it on the same or a different computing device. The computing device may serialize the data structure to send and receive data over a network. A data structure including weights of the serialized neural network may be reconstructed in the same computing device or in another computing device through deserialization. The data structure including the weight of the neural network is not limited to serialization. Furthermore, the data structure including the weights of the neural network is a data structure to increase the efficiency of computation while using the resources of the computing device to a minimum (e.g., in a nonlinear data structure, B-Tree, Trie, m-way search tree, AVL tree, Red-Black Tree). The foregoing is merely an example, and the present disclosure is not limited thereto.

The data structure may include hyper-parameters of the neural network. In addition, the data structure including the hyperparameters of the neural network may be stored in a computer-readable medium. The hyperparameter may be a variable variable by a user. Hyperparameters are, for example, learning rate, cost function, number of iterations of the learning cycle, weight initialization (e.g., setting the range of weight values to be initialized for weights), Hidden Unit The number (eg, the number of hidden layers, the number of nodes of the hidden layer) may be included. The above-described data structure is merely an example, and the present disclosure is not limited thereto.

3 is an exemplary diagram for explaining a method for providing an artificial intelligence service.

In FIG. 3 , the server 100, the edge devices 210, 230, and 250, and the coordinator module mapped to each server 100 and/or the edge device 210, 230, 250, learning result information 270, state Information 290 is shown.

According to an embodiment of the present disclosure, the coordinator module may include a module physically independent from the server 100 or the edge devices 210 , 230 , and 250 . Accordingly, the coordinator module can operate normally even when a failure occurs in the server 100 or the edge devices 210 , 230 , and 250 . The coordinator module may receive status information of the mapped server or edge device and transmit the information to a physically independent device such as another server, another edge device, or another coordinator module. There may be only one coordinator module, or a plurality of coordinator modules may exist. When only one coordinator module exists, the corresponding coordinator module may have a one-to-many mapping relationship with the server and/or all edge devices. When a plurality of coordinator modules exist, each of the server and/or edge device may have a one-to-one mapping relationship. The above-described mapping relationship is merely an example, and the present disclosure is not limited thereto. According to another embodiment of the present disclosure, the coordinator module may be a module included in the server and/or the edge device. The above-described coordinator module is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the server 100 may receive status information from the coordinator module mapped with the server 100 . The corresponding state information may be state information of the edge devices 210 , 230 , and 250 . The processor 110 may receive state information of the edge devices 210 , 230 , and 250 from a coordinator module mapped with the edge devices 210 , 230 , and 250 , respectively. In addition, the processor 110 may receive the state information of the edge devices 210 , 230 , and 250 from the coordinator module mapped with the server 100 . The coordinator module may acquire status information of the server and/or edge device through monitoring in real time. According to an embodiment of the present disclosure, the processor 110 may receive the learning result information 270 from the edge device 210 and/or the coordinator module mapped with the edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

4 is an exemplary diagram for explaining a synchronization operation performed when a failure occurs in an edge device in the process of providing an artificial intelligence service.

4 is a mapping to the server 100, the edge devices 230 and 250, the faulty edge device 310 and each server 100, the faulty edge device 310 and/or the edge devices 230 and 250 An illustrated coordinator module is shown.

According to an embodiment of the present disclosure, the server 100 may perform a first synchronization operation of transmitting a model to the edge device 310 having a failure. The first synchronization operation may include an operation of performing synchronization by transmitting (313) a model to the edge device having a failure based on state information received from the coordinator module mapped with the edge device 310 in which the failure occurred. The server 100 may receive the abnormal state information 311 from the faulty edge device 310 and/or the coordinator module mapped with the faulty edge device 310 . In this case, the server 100 may directly receive the abnormal state information 311 , or may receive the abnormal state information 311 indirectly (bypass) through the coordinator module mapped with the server 100 . The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, there may exist a case in which the version of the model stored in the faulty edge device 310 is lower than the version of the model stored in the server. The version information of the model may include information for identifying when the learned model is updated. The server 100 may compare the version of the model based on the version information of the model. For example, for the same model A, if model A was updated on 2015-01-01, model version information is model A-v1, updated: 2015-01-01, model A updated on 2019-01-01 If it is, the model version information may be model A-v3, updated date: 2019-01-01. Accordingly, the server 100 may compare the versions of the model A-v1 and the model A-v3 to determine that the model A-v3 is a higher version compared to the model A-v1. Cases in which the edge device fails may include cases in which the latest model cannot be transmitted from the server 100, cases in which the edge device itself cannot update the model, and the like. In this case, there may be a need to update the model to the latest model for an AI service that can continuously provide the latest model. Therefore, the server 100 transmits the model stored in the server 100 to at least one of the faulty edge device 310 and the mapped coordinator module or the faulty edge device 310 to cause the faulty edge device 310 to the server. It can cause updates to the model stored in . Here, an error may occur in the network unit of the edge device 310 having a failure, and direct communication with the server may not be possible. In this case, the server may transmit the model to the faulty edge device 310 through the coordinator module mapped with the faulty edge device 310 . The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, there may exist a case where the version of the model stored in the faulty edge device 310 is lower than the version of the model stored in other edge devices. The server 100 may compare the version of the model stored in each edge device based on the version information of the model. Therefore, the server 100 causes the other edge devices 230 and 250 to transfer the model stored in the other edge devices 230 and 250 to the faulty edge device 310 and the mapped coordinator module or the faulty edge device 310 . It can be transmitted to at least one to cause the faulty edge device 310 to update the model stored in another edge device. Here, an error may occur in the network unit of the edge device 310 having a failure, and direct communication with other edge devices 230 and 250 may not be possible. In this case, the server 100 may cause other edge devices 230 and 250 to transmit the model to the edge device 310 in which the failure occurs through the coordinator module mapped with the edge device 310 in which the failure occurred. Accordingly, the server 100 may cause the faulty edge device to update the model stored in the other edge devices 230 and 250 . Through this, a sudden increase in the load on the server is reduced, so that the speed at which the server transmits the model and the speed at which the server processes the task is maintained constant without abruptly decreasing. The foregoing is merely an example, and the present disclosure is not limited thereto.

5 is an exemplary diagram for explaining a synchronization operation performed when a failure occurs in a server in the process of providing an artificial intelligence service.

5 shows a failed server 100, edge devices 230 and 250, a failed edge device 310 and each server 100, a failed edge device 310 and/or edge devices 230 and 250 The coordinator module mapped to is shown.

According to an embodiment of the present disclosure, the synchronization operation may include a second synchronization operation that causes the other edge devices 230 and 250 to transmit a model to the edge device 310 where the failure occurs. The second synchronization operation causes the other edge devices 230 and 250 to transmit a model to the edge device 310 where the failure occurs based on the status information received from the coordinator module mapped with the failed server, thereby performing synchronization. may include The server 100 having a failure may include a server that cannot perform a service providing the most recently learned model. For example, the edge devices 230 and 250 may have model A1 updated on 10/10/2019 with respect to model A. On the other hand, the server 100 having a failure may store the model A2 updated on 10/10/2016 for the same model A. That is, the server 100 having a failure may not store the most recently updated model. Accordingly, the server 100 having a failure may not be able to perform a service of providing the most recently learned model A1 to other edge devices. According to another embodiment of the present disclosure, the server 100 having a failure may be in a state in which communication with other servers and edge devices is impossible due to a problem in the computing device. Also, the server 100 having a failure may be a server that cannot perform model update due to a problem in the computing device. Also, the faulty edge device may be an edge device that has lost the latest version model stored in the memory due to a problem in the computing device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the second synchronization operation causes other edge devices 230 and 250 to model the faulty edge device based on the status information received from the coordinator module mapped with the faulty server 100. may include the task of performing synchronization by causing the In this case, since a failure has occurred in the server, the server 100 may not be able to directly transmit the model to the edge device. As another example, since a failure occurs in the server, there may exist a case where the version of the model stored in the faulty edge device 310 is lower than the version of the model stored in the other edge devices 230 and 250 . The server 100 may compare the version of the model stored in each edge device based on the version information of the model. In this case, the server 100 transmits the model stored in the other edge devices 230 and 250 to at least one of the faulty edge device 310 and the mapped coordinator module or the faulty edge device 310 to the faulty edge device. may cause 310 to update with the model stored in other edge devices 230 , 250 . Through this, artificial intelligence services can be provided without interruption even when the server fails. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the third synchronization operation may include an operation of receiving a model from an edge device or a coordinator module mapped with an edge device based on state information received from a coordinator module mapped with a server with a failure. have. By performing the third synchronization operation, the server 100 having a failure may update the model to the latest version of the model. Furthermore, the server 100 in which the failure has occurred may store the latest version of the model in the edge device. Through this, it is possible to provide a continuous artificial intelligence service by normalizing the server 100 in which the failure occurs in a short time. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, in the third synchronization operation, when the version of the model stored in the edge device 230 is higher than the version of the model stored in the server 100 in which the failure occurs, the model stored in the edge device 230 is higher. The method may further include the operation of receiving from at least one of the edge device 230 or the coordinator module mapped with the edge device 230 and updating the model with the model stored in the edge device 230 . There may exist a case where the version of the model stored in the edge device 230 is higher than the version of the model stored in the server 100 in which the failure occurs. The server 100 may compare the version of the model stored in each edge device based on the version information of the model. When a failure occurs in the server 100 , the server 100 may not be able to update the model. Therefore, the version of the model stored in the server 100 in which the failure occurs may be lower than the version of the model stored in the edge device. In this case, the server 100 having a failure may receive ( 450 ) the model stored in the edge device 230 from at least one of the edge device 230 or a coordinator module mapped with the edge device 230 . In this case, the server 100 in which the failure occurs may receive the model stored in the edge device 230 through the coordinator module mapped with the server 100 in which the failure occurs. The reason is that there may exist a case where the server 100 in which the failure occurs cannot directly receive the model from the edge device 230 . The server 100 having a failure may receive the model stored in the edge device 230 and update the model. The server 100 in which the failure has occurred may transmit the updated model to the edge device 310 in which the failure has occurred after updating the model. The foregoing is merely an example, and the present disclosure is not limited thereto.

6 is an exemplary diagram illustrating a multiplexing network for providing an uninterrupted artificial intelligence service.

6 shows the server 100 and at least one lightweight edge device and at least one heavyweight edge device.

According to an embodiment of the present disclosure, the server 100 may cause the lightweight edge device to transmit the model to the faulty lightweight edge device and/or the faulty heavy edge device. In addition, the server 100 may cause the heavy edge device to transmit the model to the faulty weight edge device and/or the faulty lightweight edge device. The failed server may receive the model from the lightweight edge device and/or the heavy edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, at least one heavy edge device may be present between the server and the lightweight edge device or may be omitted. As shown in FIG. 6 , a heavy edge device 1 720 may exist between the server 100 and the lightweight edge device 2 721 . According to another embodiment of the present disclosure, at least one lightweight edge device may exist between the server 100 and the lightweight edge device 4 751 or may be omitted. According to another embodiment of the present disclosure, at least one other lightweight edge device and at least one weight edge device may exist between the server 100 and the lightweight edge device 4 751 . According to another embodiment of the present disclosure, at least one other weight edge device 2 730 may exist or may be omitted between the server 100 and the weight edge device 3 731 . Also, at least one other edge device may exist between the server and the weight edge device or may be omitted. There may be at least one lightweight edge device and at least one weight edge device between the server and the weight edge device. Through this, multiplexed connection more than tripled can be implemented in the artificial intelligence service providing system. Accordingly, the processor 110 may provide the model to the user without interruption through distributed processing even when a failure occurs in the server and/or the edge device. The foregoing is merely an example, and the present disclosure is not limited thereto.

According to an embodiment of the present disclosure, the number of edge devices existing between the server and the edge device may be defined as a hop. In this case, edge devices present in hop 0 are lightweight edge device 1 710 , weight edge device 1 720 , weight edge device 2 730 , weight edge device 4 740 , and weight edge device 5 750 . can exist. Edge devices existing in hop 0 can transmit and receive models to each other. Accordingly, the processor 110 may cause the normal 0-hop edge device to transmit the model to the faulty edge device. The one-hop edge device may be a lightweight edge device 2 721 and a heavy edge device 3 731 . In a multiplexed network for providing artificial intelligence services, hops may be n-hops. In this case, n may be any positive integer. Therefore, the processor ( ) can guarantee high availability by providing artificial intelligence services in an n-hop multiplexed network. The foregoing is merely an example, and the present disclosure is not limited thereto.

7 is a flowchart for providing an artificial intelligence service.

According to an embodiment of the present disclosure, the server 100 may transmit a model to at least one edge device to allow the edge device to train the model 510 .

According to an embodiment of the present disclosure, information on the learning result of the model may be received ( 520 ) from at least one edge device.

According to an embodiment of the present disclosure, the information on the learning result of the model may include at least one of version information of the learned model, weight of the learned model, weight change amount according to learning, or data related to the learning process of the model. .

According to an embodiment of the present disclosure, the weight of the learned model may reflect pattern information extracted from the training data set used to train the model in the edge device.

According to an embodiment of the present disclosure, the server 100 may update 530 the model based on the learning result information received from at least one edge device.

According to an embodiment of the present disclosure, the synchronization operation may include a first synchronization operation to transmit a model to a faulty edge device, a second synchronization operation to cause another edge device to transmit a model to a faulty edge device, or an edge and at least one of a third synchronization operation of receiving the model from the device.

According to an embodiment of the present disclosure, the state information may include at least one of version information of the model, meta information of learning data of the model, storage location information of the model, operation-related information of an edge device, or operation-related information of a server have.

According to an embodiment of the present disclosure, the first synchronization operation may include an operation of performing synchronization by transmitting a model to an edge device having a failure based on state information received from a coordinator module mapped with the edge device in which the failure occurred. have.

According to an embodiment of the present disclosure, in the first synchronization operation, when the version of the model stored in the faulty edge device is lower than the version of the model stored in the server, the model stored in the server is mapped to the faulty edge device by the coordinator. sending to at least one of the module or the faulty edge device to cause the faulty edge device to update the model stored in the server; Alternatively, if the version of the model stored in the faulty edge device is lower than the version of the model stored in the other edge device, the other edge device can convert the model stored in the other edge device to the coordinator module mapped with the faulty edge device or the faulty edge. sending to at least one of the devices to cause the failed edge device to update the model stored in the other edge device; may include at least one of

According to an embodiment of the present disclosure, the second synchronization operation is performed by causing another edge device to transmit a model to the faulty edge device based on the status information received from the coordinator module mapped with the failed server. may include tasks.

According to an embodiment of the present disclosure, in the second synchronization operation, when the version of the model stored in the faulty edge device is lower than the version of the model stored in the other edge device, the second synchronization operation converts the model stored in the other edge device to the faulty edge device. and sending to at least one of the mapped coordinator module or the faulty edge device to cause the faulty edge device to update the model stored in the other edge device.

According to an embodiment of the present disclosure, the third synchronization operation may include an operation of receiving a model from an edge device or a coordinator module mapped with an edge device based on state information received from a coordinator module mapped with a server with a failure. can

According to an embodiment of the present disclosure, in the third synchronization operation, when the version of the model stored in the edge device is higher than the version of the model stored in the server in which the failure occurs, the model stored in the edge device is mapped to the edge device or the edge device. The method may further include updating a model stored in the edge device by receiving it from at least one of the coordinated coordinator modules.

According to an embodiment of the present disclosure, an edge device includes: a weight edge device capable of performing model training and model transmission; or a lightweight edge device capable of performing model transmission; may include at least one of

8 is a block diagram illustrating a module for providing an artificial intelligence service.

A method for providing an artificial intelligence service according to an embodiment of the present disclosure may be implemented by the following modules. a module 610 for sending a model to at least one edge device to cause the edge device to train the model; a module 620 for receiving model learning result information from at least one edge device; and a module 630 for updating the model based on the learning result information received from at least one edge device.

In an alternative embodiment for providing an artificial intelligence service, the module 610 for sending a model to at least one edge device so that the edge device learns the model is based on the model request signal received from the edge device. It may include an operation of transmitting the model requested by the user to the edge device.

In an alternative embodiment for providing an artificial intelligence service, the module 630 for updating the model based on the learning result information received from the at least one edge device may include a weight of the learned model received from the at least one edge device. a module for updating the model based on at least some of the; a module for updating the model based on at least a portion of the two or more learned models received from the two or more edge devices; or a module for updating version information of the model; may include at least one of

In an alternative embodiment for providing an artificial intelligence service, the module may further include a module for performing a synchronization operation based on the state information. The synchronization operation may include updating the model in order to perform a service in which the server or at least one edge device provides the same model to the user.

In an alternative embodiment for providing an artificial intelligence service, the module for transmitting the model may include a module for transmitting to the user terminal via at least one of at least one heavy edge device or at least one lightweight edge device. have.

According to an embodiment of the present disclosure, a module for providing an artificial intelligence service may be implemented by means, circuitry or logic for implementing a computing device. Those skilled in the art will further appreciate that the various illustrative logical blocks, configurations, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be combined with electronic hardware, computer software, or combinations of both. It should be recognized that it can be implemented as To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

9 is a simplified, general schematic diagram of an exemplary computing environment in which embodiments of the present disclosure may be implemented.

Although the present disclosure has been described above as being generally capable of being implemented by a computing device, those skilled in the art will appreciate that the present disclosure is a combination of hardware and software and/or in combination with computer-executable instructions and/or other program modules that may be executed on one or more computers. It will be appreciated that it can be implemented as a combination.

Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will appreciate that the methods of the present disclosure are suitable for single-processor or multiprocessor computer systems, minicomputers, mainframe computers as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, etc. (each of which It will be appreciated that other computer system configurations may be implemented, including those that may operate in connection with one or more associated devices.

The described embodiments of the present disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

Computers typically include a variety of computer-readable media. Any medium accessible by a computer can be a computer-readable medium, and such computer-readable media includes volatile and nonvolatile media, transitory and non-transitory media, removable and non-transitory media. including removable media. By way of example, and not limitation, computer-readable media may include computer-readable storage media and computer-readable transmission media. Computer-readable storage media includes volatile and non-volatile media, transitory and non-transitory media, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. includes media. A computer-readable storage medium may be RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital video disk (DVD) or other optical disk storage device, magnetic cassette, magnetic tape, magnetic disk storage device, or other magnetic storage device. device, or any other medium that can be accessed by a computer and used to store the desired information.

Computer readable transmission media typically embodies computer readable instructions, data structures, program modules or other data, etc. in a modulated data signal such as a carrier wave or other transport mechanism, and Includes any information delivery medium. The term modulated data signal means a signal in which one or more of the characteristics of the signal is set or changed to encode information in the signal. By way of example, and not limitation, computer-readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above are also intended to be included within the scope of computer-readable transmission media.

An example environment 1100 implementing various aspects of the disclosure is shown including a computer 1102 , the computer 1102 including a processing unit 1104 , a system memory 1106 , and a system bus 1108 . do. The system bus 1108 couples system components, including but not limited to system memory 1106 , to the processing device 1104 . The processing device 1104 may be any of a variety of commercially available processors. Dual processor and other multiprocessor architectures may also be used as processing unit 1104 .

The system bus 1108 may be any of several types of bus structures that may be further interconnected to a memory bus, a peripheral bus, and a local bus using any of a variety of commercial bus architectures. System memory 1106 includes read only memory (ROM) 1110 and random access memory (RAM) 1112 . A basic input/output system (BIOS) is stored in non-volatile memory 1110, such as ROM, EPROM, EEPROM, etc., which is the basic input/output system (BIOS) that helps transfer information between components within computer 1102, such as during startup. contains routines. RAM 1112 may also include high-speed RAM, such as static RAM, for caching data.

The computer 1102 may also be configured with an internal hard disk drive (HDD) 1114 (eg, EIDE, SATA) - this internal hard disk drive 1114 may also be configured for external use within a suitable chassis (not shown). Yes-, magnetic floppy disk drive (FDD) 1116 (eg, for reading from or writing to removable diskette 1118), and optical disk drive 1120 (eg, CD-ROM) for reading from, or writing to, disk 1122, or other high capacity optical media such as DVD. The hard disk drive 1114 , the magnetic disk drive 1116 , and the optical disk drive 1120 are connected to the system bus 1108 by the hard disk drive interface 1124 , the magnetic disk drive interface 1126 , and the optical drive interface 1128 , respectively. ) can be connected to The interface 1124 for external drive implementation includes at least one or both of Universal Serial Bus (USB) and IEEE 1394 interface technologies.

These drives and their associated computer-readable media provide non-volatile storage of data, data structures, computer-executable instructions, and the like. In the case of computer 1102, drives and media correspond to storing any data in a suitable digital format. Although the description of computer readable media above refers to HDDs, removable magnetic disks, and removable optical media such as CDs or DVDs, those skilled in the art will use zip drives, magnetic cassettes, flash memory cards, cartridges, etc. It will be appreciated that other tangible computer-readable media such as etc. may also be used in the exemplary operating environment and any such media may include computer-executable instructions for performing the methods of the present disclosure.

A number of program modules may be stored in the drive and RAM 1112 , including an operating system 1130 , one or more application programs 1132 , other program modules 1134 , and program data 1136 . All or portions of the operating system, applications, modules, and/or data may also be cached in RAM 1112 . It will be appreciated that the present disclosure may be implemented in various commercially available operating systems or combinations of operating systems.

A user may enter commands and information into the computer 1102 via one or more wired/wireless input devices, for example, a pointing device such as a keyboard 1138 and a mouse 1140 . Other input devices (not shown) may include a microphone, IR remote control, joystick, game pad, stylus pen, touch screen, and the like. Although these and other input devices are often connected to the processing unit 1104 through an input device interface 1142 that is connected to the system bus 1108, parallel ports, IEEE 1394 serial ports, game ports, USB ports, IR interfaces, It may be connected by other interfaces, etc.

A monitor 1144 or other type of display device is also coupled to the system bus 1108 via an interface, such as a video adapter 1146 . In addition to the monitor 1144, the computer typically includes other peripheral output devices (not shown), such as speakers, printers, and the like.

Computer 1102 may operate in a networked environment using logical connections to one or more remote computers, such as remote computer(s) 1148 via wired and/or wireless communications. Remote computer(s) 1148 may be workstations, computing device computers, routers, personal computers, portable computers, microprocessor-based entertainment devices, peer devices, or other common network nodes, and are typically connected to computer 1102 . Although it includes many or all of the components described for it, only memory storage device 1150 is shown for simplicity. The logical connections shown include wired/wireless connections to a local area network (LAN) 1152 and/or a larger network, eg, a wide area network (WAN) 1154 . Such LAN and WAN networking environments are common in offices and companies, and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to a worldwide computer network, for example, the Internet.

When used in a LAN networking environment, the computer 1102 is connected to the local network 1152 through a wired and/or wireless communication network interface or adapter 1156 . Adapter 1156 may facilitate wired or wireless communication to LAN 1152 , which also includes a wireless access point installed therein for communicating with wireless adapter 1156 . When used in a WAN networking environment, the computer 1102 may include a modem 1158, be connected to a communication computing device on the WAN 1154, or establish communications over the WAN 1154, such as over the Internet. have other means. A modem 1158 , which may be internal or external and a wired or wireless device, is coupled to the system bus 1108 via a serial port interface 1142 . In a networked environment, program modules described for computer 1102 , or portions thereof, may be stored in remote memory/storage device 1150 . It will be appreciated that the network connections shown are exemplary and other means of establishing a communication link between the computers may be used.

Computer 1102 may be associated with any wireless device or object that is deployed and operates in wireless communication, for example, printers, scanners, desktop and/or portable computers, portable data assistants (PDAs), communications satellites, wireless detectable tags. It operates to communicate with any device or place, and phone. This includes at least Wi-Fi and Bluetooth wireless technologies. Accordingly, the communication may be a predefined structure as in a conventional network or may simply be an ad hoc communication between at least two devices.

Wi-Fi (Wireless Fidelity) makes it possible to connect to the Internet, etc. without a wire. Wi-Fi is a wireless technology such as cell phones that allows these devices, eg, computers, to transmit and receive data indoors and outdoors, ie anywhere within range of a base station. Wi-Fi networks use a radio technology called IEEE 802.11 (a, b, g, etc.) to provide secure, reliable, and high-speed wireless connections. Wi-Fi can be used to connect computers to each other, to the Internet, and to wired networks (using IEEE 802.3 or Ethernet). Wi-Fi networks may operate in unlicensed 2.4 and 5 GHz radio bands, for example, at 11 Mbps (802.11a) or 54 Mbps (802.11b) data rates, or in products that include both bands (dual band). .

One of ordinary skill in the art of this disclosure will understand that information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, instructions, information, signals, bits, symbols and chips that may be referenced in the above description are voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields particles or particles, or any combination thereof.

A person of ordinary skill in the art of the present disclosure will recognize that the various illustrative logical blocks, modules, processors, means, circuits and algorithm steps described in connection with the embodiments disclosed herein include electronic hardware, (convenience It will be understood that it may be implemented by various forms of program or design code (referred to herein as software) or a combination of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. A person skilled in the art of the present disclosure may implement the described functionality in various ways for each specific application, but such implementation decisions should not be interpreted as a departure from the scope of the present disclosure.

The various embodiments presented herein may be implemented as methods, apparatus, or articles of manufacture using standard programming and/or engineering techniques. The term article of manufacture includes a computer program, carrier, or media accessible from any computer-readable storage device. For example, computer-readable storage media include magnetic storage devices (eg, hard disks, floppy disks, magnetic strips, etc.), optical disks (eg, CDs, DVDs, etc.), smart cards, and flash drives. memory devices (eg, EEPROMs, cards, sticks, key drives, etc.). Also, various storage media presented herein include one or more devices and/or other machine-readable media for storing information.

It is understood that the specific order or hierarchy of steps in the presented processes is an example of exemplary approaches. Based on design priorities, it is understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present disclosure. The appended method claims present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present disclosure is not intended to be limited to the embodiments presented herein, but is to be construed in the widest scope consistent with the principles and novel features presented herein.

Claims (18)

A computer program stored in a computer readable storage medium, wherein the computer program, when executed on one or more processors, performs the following operations for providing an artificial intelligence service, the operations comprising:
transmitting a model to at least one edge device so that the edge device learns the model;
receiving information about a learning result of the model from at least one edge device; and
updating the model based on the learning result information received from at least one edge device;
comprising,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The operation of sending a model to the at least one edge device so that the edge device learns the model includes:
transmitting a model requested by a user to the edge device based on a model request signal received from the edge device;
comprising,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The learning result information of the model is,
At least one of the version information of the learned model, the weight of the learned model, the weight change according to learning, or data related to the learning process of the model
A computer program stored in a computer-readable storage medium.
4. The method of claim 3,
The weight of the learned model is,
Reflecting the pattern information extracted from the training data set used to train the model on the edge device,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
Updating the model based on the learning result information received from the at least one edge device includes:
updating the model based on at least a part of weights of the learned model received from the at least one edge device;
updating the model based on at least a portion of two or more learned models received from two or more edge devices; or
updating version information of the model;
comprising at least one of
A computer program stored in a computer-readable storage medium.
The method of claim 1,
performing a synchronization operation based on the state information;
further comprising,
The synchronization operation is
Updating the model in order for the server or at least one edge device to provide the same model to the user
comprising,
A computer program stored in a computer-readable storage medium.
7. The method of claim 6,
The synchronization operation is
The first synchronization operation to send the model to the failed edge device;
a second synchronization operation that causes another edge device to send a model to the failed edge device; or
3rd sync operation to receive model from edge device
comprising at least one of
A computer program stored in a computer-readable storage medium.
7. The method of claim 6,
The status information is
Model version information, model learning data meta information, model storage location information, edge device operation-related information, or server operation-related information
comprising at least one of
A computer program stored in a computer-readable storage medium.
8. The method of claim 7,
The first synchronization operation is
A task of performing synchronization by transmitting a model to the faulty edge device based on the status information received from the faulty edge device and the mapped coordinator module
comprising,
A computer program stored in a computer-readable storage medium.
10. The method of claim 9,
The first synchronization operation is
When the version of the model stored in the faulty edge device is lower than the version of the model stored in the server, the model stored in the server is transmitted to at least one of a coordinator module mapped with the faulty edge device or the faulty edge device to cause the faulty edge device to update the model stored in the server; or
When the version of the model stored in the faulty edge device is lower than the version of the model stored in other edge devices, the coordinator module that maps the model stored in the other edge device to the faulty edge device or sending to at least one of the faulty edge devices to cause the faulty edge device to update the model stored in the other edge device;
comprising at least one of
A computer program stored in a computer-readable storage medium.
8. The method of claim 7,
The second synchronization operation is
Synchronization by causing another edge device to transmit a model to the faulty edge device based on the status information received from the failed server and the mapped coordinator module
comprising,
A computer program stored in a computer-readable storage medium.
12. The method of claim 11,
The second synchronization operation is
If the version of the model stored in the faulty edge device is lower than the version of the model stored in other edge devices, the model stored in the other edge device is mapped to the faulty edge device and the coordinator module or the faulty edge device. Sending to at least one to cause the failed edge device to update the model stored in the other edge device
comprising,
A computer program stored in a computer-readable storage medium.
8. The method of claim 7,
The third synchronization operation is
The operation of receiving a model from an edge device or a coordinator module mapped with an edge device based on the status information received from the coordinator module mapped with the server in which the failure occurred
comprising,
A computer program stored in a computer-readable storage medium.
14. The method of claim 13,
The third synchronization operation is
When the version of the model stored in the edge device is higher than the version of the model stored in the server in which the failure occurs, the model stored in the edge device is received from at least one of the edge device or a coordinator module mapped with the edge device. Updating the model stored on the edge device
further comprising,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The edge device is
a weight edge device capable of learning the model and transmitting the model; or
a lightweight edge device capable of performing the transmission of the model;
comprising at least one of
A computer program stored in a computer-readable storage medium.
16. The method of claim 15,
The transmission of the model is
An operation of being transmitted to a user terminal via at least one of the at least one weight edge device or the at least one lightweight edge device
comprising,
A computer program stored in a computer-readable storage medium.
A method for providing an artificial intelligence service,
transmitting the model to at least one edge device to cause the edge device to train the model;
Receiving learning result information of the model from at least one edge device; and
updating the model based on the learning result information received from at least one edge device;
containing,
A method for providing artificial intelligence services.
As a server for providing artificial intelligence services,
one or more processors; and
a memory storing instructions executable by the one or more processors;
including,
The one or more processors,
transmit the model to at least one edge device to cause the edge device to train the model,
Receiving learning result information of the model from at least one edge device, and
Updating the model based on the learning result information received from at least one edge device,
server.

Images