KR102292703B1 - Method for transfer learning based on feature set information - Google Patents

Abstract

According to one embodiment of the present disclosure, a computer program stored in a computer-readable storage medium is disclosed. The computer program includes instructions for allowing at least one processor to perform transfer learning on a neural network, in which the instructions include: analyzing meta information of one or more pre-learning models and meta information of a learning target model, in which the meta information includes feature set information and role information of each of the one or more pre-learning models, and feature set information and role information of the learning target model; determining a seed model among the one or more pre-learning models based on an analysis result of the feature set information and the role information; and performing transfer learning on the learning target model by using weight information of the seed model. Accordingly, a transfer learning method for a learning target model is provided.

Description

Transfer learning method based on feature set information

The present invention relates to a method for training a neural network related to the operation of equipment in a manufacturing field or the like.

Anomaly detection algorithm using a conventional neural network is generally applicable to all fields. However, in some manufacturing industries (e.g. robotic arms, semiconductors, batteries, etc.), equipment in many manufacturing industries is used without being customized for a specific process.

In the algorithm used in these off-the-shelf devices, it may be advantageous to perform transfer learning from a model pre-trained with similar devices.

Therefore, there may be a demand in the art for a method of performing transfer learning from a previously learned model.

The present disclosure has been devised in response to the above background art, and an object of the present disclosure is to provide a transfer learning method based on meta information of a neural network.

Disclosed is a computer program stored in a computer-readable storage medium for solving the above problems. The computer program includes instructions for causing one or more processors to perform transfer learning on a neural network, the instructions comprising: analyzing meta-information of one or more pre-learning models and meta-information of a learning target model - the meta-information the information includes feature set information and role information of each of the one or more prior learning models, and feature set information and role information of the learning target model; determining a seed model among the one or more pre-learning models based on the analysis result of the meta information; and performing transfer learning on the learning target model using weight information of the seed model. may include.

In addition, the feature set information may be information on a shape of an input vector of each of the one or more pre-learning models and the learning target model.

In addition, the role information includes at least one of a process, a machine device, or a process object to which at least one of the pre-learning model or the learning target model is related, and hierarchical information between the process, the machine device, and the process object. can do.

In addition, the determining of the seed model among the one or more pre-learning models based on the analysis result of the meta information may include: identifying a first pre-learning model whose feature set information satisfies a preset criterion; and determining the seed model to be used for transfer learning based on a comparison result of the role information of the first pre-learning model and the learning target model. may include.

In addition, the determining of the seed model to be used for transfer learning based on a comparison result of the role information of the first pre-learning model and the learning target model may include features of the first pre-learning model and the learning target model. determining an all-data learning model as the seed model when set information is matched and role information of the first pre-learning model and the learning target model are different; Alternatively, when the feature set information of the first pre-training model and the learning target model do not match and the role information of the first pre-training model and the learning target model are similar, the first pre-learning model is used as the seed model determining as; may include.

In addition, the performing transfer learning on the learning target model by using the weight information of the seed model may include applying weight information of the seed model related to a common feature set of the seed model and the learning target model to the learning target. performing transfer learning by applying it to the model; may include.

In addition, the performing transfer learning by applying the weight information of the seed model to the learning target model includes: the seed model and the learning target model from the feature set information of the seed model and the feature set information of the learning target model extracting a common feature set of ; and applying weight information of some layers or all layers related to the common feature set of the seed model as initial weight information of the learning target model. may include.

In addition, the step of applying weight information of some layers or all layers related to the common feature set of the seed model as the initial weight information of the learning target model includes: feature set information of the seed model and feature set information of the learning target model based on , determining a size of a common feature set between the seed model and the learning target model; determining the number of layers of the seed model to be applied as initial weight information of the learning target model based on the size of the common feature set; may include.

Disclosed is a method for performing transfer learning on a neural network in order to solve the above-described problem. The method includes: analyzing meta-information of one or more pre-learning models and meta-information of a learning target model, wherein the meta information includes feature set information and role information of each of the one or more pre-learning models and features of the learning target model Includes set information and role information-; determining a seed model among the one or more pre-learning models based on an analysis result of the feature set information and the role information; and performing transfer learning on the learning target model using weight information of the seed model. may include.

Disclosed is a computing device for performing transfer learning on a neural network in order to solve the above-described problem. The computing device may include: a memory; and a processor; including, wherein the processor analyzes meta-information of one or more pre-learning models and meta-information of a learning target model, and the meta information includes feature set information and role information of each of the one or more pre-learning models and the learning target. including feature set information and role information of the model; based on a result of reviewing the feature set information and the role information, determine a seed model among the one or more pre-trained models, and weight the seed model Transfer learning may be performed on the learning target model using the information.

Disclosed is a computer-readable recording medium storing a data structure corresponding to a parameter of a neural network that is at least partially updated in a learning process for solving the above-described problem. The operation of the neural network is based at least in part on the parameters, and the learning process may include reviewing one or more pre-trained models and feature set information and role information of the learning target model; determining a seed model among the one or more pre-learning models based on a result of reviewing the feature set information and the role information; and performing transfer learning on the learning target model using weight information of the seed model. may include.

The present disclosure may provide a transfer learning method for a learning target model.

1 is a block diagram of a computing device for a processor to measure a bone age of a whole bone image according to an embodiment of the present disclosure.
2 is a schematic diagram illustrating a network function according to an embodiment of the present disclosure;
3 is a schematic diagram illustrating feature set information according to an embodiment of the present disclosure.
4 is a schematic diagram illustrating role information according to an embodiment of the present disclosure.
5 is a schematic diagram illustrating an initial state of a network in which transfer learning is performed according to an embodiment of the present disclosure.
6 is a schematic diagram illustrating an initial state of a network in which transfer learning is performed according to an embodiment of the present disclosure.
7 is a flowchart illustrating a process in which a processor performs transfer learning according to an embodiment of the present disclosure.
8 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 apparent 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 computing device and the computing device may be a component. 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.

In the present disclosure, a network function and an artificial neural network and a neural network may be used interchangeably.

1 is a block diagram of a computing device according to an embodiment of the present disclosure.

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

The computing device 100 may include a processor 110 , a memory 120 , and a network unit (not shown).

The processor 110 may include one or more cores, and a central processing unit (CPU) of a computing device, a general purpose graphics processing unit (GPGPU), and a tensor processing unit (TPU). unit) and the like, and may include a processor for data analysis and deep learning. The processor 110 may read a computer program stored in a memory and perform data processing for machine learning according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the processor 110 may perform an operation for learning the neural network. The processor 110 for learning of the neural network, such as processing input data for learning in deep learning (DL), extracting features from the input data, calculating an error, updating the weight of the neural network using backpropagation calculations can be performed. At least one of a CPU, a GPGPU, and a TPU of the processor 110 may process learning of a network function. For example, the CPU and the GPGPU can process learning of a network function and data classification using the network function together. In addition, in an embodiment of the present disclosure, learning of a network function and data classification using the network function may be processed by using the processors of a plurality of computing devices together. In addition, the computer program executed in the computing device according to an embodiment of the present disclosure may be a CPU, GPGPU, or TPU executable program.

According to an embodiment of the present disclosure, the memory 120 may store any type of information generated or determined by the processor 110 and any type of information received by the network unit.

According to an embodiment of the present disclosure, the memory 120 is a flash memory type, a hard disk type, a multimedia card micro type, or a card type memory (eg, SD or XD memory, etc.), Random Access Memory (RAM), Static Random Access Memory (SRAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read (PROM) -Only Memory), a magnetic memory, a magnetic disk, and an optical disk may include at least one type of storage medium. The computing device 100 may operate in relation to a web storage that performs a storage function of the memory 120 on the Internet. The description of the above-described memory is only an example, and the present disclosure is not limited thereto.

A network unit according to an embodiment of the present disclosure: Public Switched Telephone Network (PSTN), x Digital Subscriber Line (xDSL), Rate Adaptive DSL (RADSL), Multi Rate DSL (MDSL), Very High Speed DSL (VDSL) ), a variety of wired communication systems such as Universal Asymmetric DSL (UADSL), High Bit Rate DSL (HDSL), and Local Area Network (LAN) can be used.

In addition, the network unit presented in this specification (Code Division Multi Access), TDMA (Time Division Multi Access), FDMA (Frequency Division Multi Access), OFDMA (Orthogonal Frequency Division Multi Access), SC-FDMA (Single Carrier-FDMA) ) and other systems may be used.

In the present disclosure, the network unit may be configured regardless of its communication mode, such as wired and wireless, and may include various communication networks such as a personal area network (PAN) and a wide area network (WAN). In addition, the network may be a well-known World Wide Web (WWW), and may use a wireless transmission technology used for short-range communication, such as infrared (IrDA) or Bluetooth (Bluetooth).

The techniques described herein may be used in the networks mentioned above, as well as in other networks.

2 is a schematic diagram illustrating a network function according to an embodiment of the present disclosure;

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 data of the output node may be determined based on data input to the input node. Here, a link 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 of the links, the two neural networks may be recognized as different from each other.

A neural network may consist of a set of one or more nodes. A subset of nodes constituting the neural network may constitute a layer. 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 traversed 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 refer to 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 do not have other input nodes connected by a link. 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. In addition, the hidden node may mean 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 progresses from the input layer 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 input layer progresses 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 (for example, 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). machine), a deep belief network (DBN), a Q network, a U network, a Siamese network, a Generative Adversarial Network (GAN), 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 autoencoder. 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 of 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 (symmetrically with the input layer). The auto-encoder can perform non-linear dimensionality reduction. The number of input layers and output layers may correspond to a dimension after preprocessing the 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 between the encoder and the 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 using at least one of supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning. Learning of the neural network may be a process of applying knowledge for the neural network to perform a specific operation to the neural network.

A neural network can be trained in a way that minimizes output errors. In the training of a neural network, iteratively inputs the training data to the neural network, calculates the output of the neural network and the target error for the training data, and calculates the error of the neural network from the output layer of the neural network to the input layer in the direction to reduce the error. It is a process of updating the weight of each node in 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, the learning data in the case of teacher learning regarding 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, in the early stage of training of a neural network, a high learning rate can be used to enable the neural network to quickly acquire a certain level of performance, thereby increasing efficiency, and using a low learning rate at the end of learning can increase accuracy.

In the training of neural networks, in general, the training data may be a subset of real data (that is, data to be processed using the trained neural network), and thus the error on the training data is reduced, but the error on the real data is reduced. 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 it sees 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 the training data, regularization, and dropout that deactivate some of the nodes of the network in the process of learning, and the use of a batch normalization layer are applied. can

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 user-defined data elements. Physical relationships between data elements may include actual relationships between data elements physically stored on a computer-readable storage medium (eg, persistent storage). 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 resources 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, also include preprocessed data for processing by the neural network, 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 the neural network. It may include a loss function for learning of . 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 preprocessed data for processing by the neural network, 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 the neural network It may be configured to include all or any combination thereof, such as a loss function for learning 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 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. A data value output from the output node may be determined based on the weight. 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 the 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 weights 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., 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 subject to weight initialization), 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 a schematic diagram illustrating feature set information according to an embodiment of the present disclosure.

In the present disclosure, meta information may be matched to a pre-learning model and a learning target model. Such meta information includes information about the types of input data and output data of the pre-training model and the learning target model, the learning method of the pre-learning model and the training target model, and the process and process in which the pre-training model and the training target model are used. It may include information about the object, the pre-learning model, and the machine device to which the learning target model is related. In particular, in the present disclosure, meta information may include feature set information and role information of the pre-learning model and the learning target model. The feature set information and role information will be described later in detail with reference to FIGS. 3 and 4 .

In the present disclosure, the feature set 200 may be input data or input vectors of the pre-training model and the learning target model. For example, the feature set 200 may be input data or input vectors for a prior learning model, a learning target model, or an all-data learning model. Such a feature set 200 may relate to an entire input vector. Also, this feature set 200 may be composed of one or more sub-feature sets. In this case, each of the sub-feature sets may be associated with a part of the input vector. For example, when the feature set 200 is a set of sub-input data recognized through a plurality of sensors, each of the sub-feature sets may correspond to any one of the plurality of sensors. That is, the features included in the sub-feature set may represent data sensed by any one of the plurality of sensors. In an embodiment of the present disclosure, the feature set may include all kinds of data obtained in an industrial field. For example, the feature set may include operation parameters of a device for production of a product in a production process of a product, sensor data obtained by operation of the device, and the like. For example, the temperature setting of equipment in a specific process, the wavelength of the laser in the case of a process using a laser, etc. may be included in the feature set processed in the present disclosure. Also, for example, it is assumed that an image is taken of the side of the car in order to confirm the work result of the side of the car, and that the photographing is performed by three sensors. In this case, the feature set 200 may include all of the sub-feature sets recognized by each sensor. In this case, each sub-feature set may correspond to partial photographing data of the side of the vehicle recognized by each sensor.

Furthermore, the feature set information according to the present disclosure may be information on the shape of input data or input vectors of one or more pre-learning models and each of the learning target models. That is, the feature set information may include information on the size of the feature set that is the input vector, the type of the feature set that is the input vector (image or voice, etc.). That is, the feature set information may be referred to as meta information about the input vector included in the corresponding feature set. Accordingly, when a plurality of feature sets exist, the processor 110 according to the present disclosure may determine the similarity of the feature sets between one or more pre-trained pre-learning models and the learning target model based on the feature set information. Based on this, the processor 110 may determine which pre-learning model to use to perform transfer learning on the learning target model. Furthermore, based on the similarity of the feature sets, the processor 110 may determine the number of layers of the pre-learning model from which weight information is to be obtained during transfer learning. Details on this will be described later.

In the present disclosure, the first feature set 210 and the second feature set 220 may be input data or input vectors of the prior learning model and the learning target model. In an embodiment, the first feature set 210 may be input data or an input vector for the prior learning model, and the second feature set 220 may be input data or an input vector for the learning target model. 3 , the first feature set 210 and the second feature set 220 may be a subset of the feature set 200 . Also, the first feature set 210 and the second feature set 220 may include input vectors corresponding to the common feature set 230 . The first feature set 210 and the second feature set 220 may also include one or more sub-feature sets like the feature set 200 . Accordingly, the first feature set 210 and the second feature set 220 may correspond to one or more partial regions of the entire image or may correspond to information sensed by one or more sensors.

Alternatively, the first feature set 210 and the second feature set 220 may be related to a specific process, a mechanical device performing the process, or a result of the process. For example, the first feature set 210 and the second feature set 220 may correspond to result data (eg, a photograph of a product after painting) of a painting operation on different products. In an embodiment, the neural network model in which the first feature set 210 and the second feature set 220 are used as input data may be a model for anomalous detection of a process. For example, the first feature set 210 and the second feature set 220 may be input data of an artificial neural network model that detects an anomaly of a process based on sensor data for a painting operation of the robot.

3 , in the present disclosure, the common feature set 230 may be a subset of the feature set 200 shared by the first feature set 210 and the second feature set 220 . For example, as described above, the first feature set 210 and the second feature set 220 may be constructed using data input from one or more partial regions or one or more sensors. That is, when the feature set 200 is sensor data on the motion of the robot in the painting process of the car side, the first feature set 210 is the operation of the robot in the painting process for the driver's door or the passenger's door on the car side. It may be sensor data for the , and the second feature set 220 may be sensor data for the operation of the robot in the painting process for the rear seat door on the side of the vehicle. In this case, the first set of features may include data from the first to third sensors of the robot for painting the side of the car, and the second set of features may include the third through the fifth sensors for painting the side of the vehicle. It may include data from sensors. In this case, the common feature set 230 may be data input from the third sensor.

In an embodiment of the present disclosure, the processor 110 may determine the seed model based on the analysis result of meta information among one or more pre-learning models. In the present disclosure, a seed model may be a model that is a basis for performing transfer learning on a learning target model. To this end, the processor 110 may determine weight information of the seed model as weight information of the learning target model when performing transfer learning on the learning target model. In one embodiment, the determination of the seed model identifies the first pre-learning model in which the feature set information satisfies a preset criterion, and based on the comparison result of the role information of the first pre-learning model and the learning target model, This can be done by determining the seed model to be used for transfer learning. In this case, the preset criterion for identifying the first pre-learning model may be a degree of similarity between the feature set of the first pre-learning model and the learning target model. That is, the number of features included in the common feature set extracted between the first pre-learning model and the learning target model is a certain number or more, or the ratio of the number of features included in the common feature set to the number of features in the first pre-learning model is The preset criterion may be satisfied when a predetermined value or more or a ratio of the number of features included in the common feature set to the number of features in the learning target model is greater than or equal to a predetermined value. For example, when the feature set of the first pre-learning model and the learning target model match, the processor 110 determines the first pre-learning model as a seed model of transfer learning by reviewing role information of the first pre-learning model Alternatively, the all-data learning model corresponding to the upper hierarchical information of the first pre-learning model may be determined as the seed model of transfer learning. When the feature set of the first pre-learning model and the learning target model do not match, the processor 110 determines the first pre-learning model as the seed model of the transfer learning model by reviewing the role information of the first pre-learning model, or Alternatively, it may be determined not to perform transfer learning using the first pre-learning model. A detailed description thereof will be described later with reference to FIGS. 4 to 7 .

As described above, the pre-learning model is determined using the feature set information representing the feature set 200 of the pre-learning model and the learning target model, and the initial training target model is formed using a network related to the common feature set 230 . By configuring the state, unnecessary information may not be brought to the learning target model among the knowledge contained in the pre-learning model. Also, even if the feature sets of the learning target model and the pre-learning model do not match as a whole, the knowledge contained in the pre-learning model can be transferred to the learning target model. Therefore, transfer learning for the learning target model can be performed more effectively.

4 is a schematic diagram illustrating role information according to an embodiment of the present disclosure.

In the present disclosure, role information may be information about a process, a machine device, and a task performed by the process or the machine device related to the pre-learning model or the learning target model. For example, assuming that the pre-learning model or the learning target model according to the present disclosure is an artificial neural network model that determines whether the painting result for the car model 1 is anonymous, the role information of the pre-learning model or the learning target model is It may be "painting process for car model 1". In the present disclosure, such a pre-learning model or a learning target model may be separately trained according to the types of input data and output data of the corresponding model and according to the hierarchy of role information. For example, assuming that the pre-training model or the learning target model is an artificial neural network model that determines whether the task result of the car model 1 is annomalous, the role information of the pre-training model or the learning target model is "car model 1 "It could be Alternatively, assuming that the pre-learning model or the learning target model is an artificial neural network model that determines whether the task result of the robot arm model A is annomalous, the role information of the pre-training model or the learning target model is "robot arm model A "It could be

As described above, each pre-learning model or learning target model may be trained using different data sets according to the hierarchy of role information. For example, when the role information is "painting process of car model 1", a pre-learning model or a learning target model corresponding to the role information may be trained using result data of the painting process of car model 1. Alternatively, the pre-learning model or the learning target model when the role information is “car model 1” may be trained using both the results of the painting process of the car model 1 and the welding process of the car model 1. In addition, when the role information is “robot arm model A”, a pre-training model or a learning target model corresponding thereto may be trained using both the vehicle model 1 and the result data of the painting process and the welding process of the vehicle model 1.

In the present disclosure, role information of one or more pre-learning models and learning target models may include hierarchical information 300 corresponding to the role information. As shown in FIG. 4 , the role information has a hierarchy of robot arm model - car model - process order. Specifically, the role information of the robot arm model may be hierarchical information 300 of level 1, the role information of the car model may be hierarchical information 300 of level 2, and the role information for the process may be hierarchical information of level 3 It may be information 300 . The processor 110 according to the present disclosure may determine the seed model based on the hierarchical information 300 . Specifically, it is assumed that the learning target model has role information of the painting process of car model 1, and the pre-training model has role information of the welding process of car model 1. Since the feature sets of the pre-training model and the learning target model are the same, but the role information is different, the processor 110 selects a pre-learning model having the role information of car model 1, which is a higher hierarchy, and selects the "welding process of car model 1". It can be determined as the seed model for transfer learning of the learning target model for ".

As described above, the role information may include at least one of machine device information 310 , process object information 320 , and process information 330 . For example, when the transfer learning method according to the present disclosure is for anomaly detection of a result of a manufacturing process, the machine device information 310 may mean a machine device that performs the manufacturing process. Also, the process object information 320 may refer to an object on which the mechanical device performs a manufacturing process or an object after a production process including the manufacturing process is finally performed. In addition, the process information 330 may mean information about an individual process being performed by the mechanical device. Since the machine device information 310 , the process object information 320 , and the process information 330 are merely examples of components for the role information, the role information may include additional other information. In addition, since the hierarchical information 300 between the elements of the role information may be configured differently from that shown in FIG. 4 , it should not be limited to the exemplary content shown in FIG. 4 and the description of the invention. .

In an embodiment, the processor 110 may determine the seed model to be used for transfer learning based on a comparison result of role information between the pre-learning model and the learning target model. Specifically, when the feature set information of the first pre-learning model and the learning target model match, and the role information of the first pre-learning model and the learning target model are different, the processor 110 selects the all-data learning model. It can be determined by the seed model. Referring to FIG. 4 , when the pre-learning model has role information of the painting process for car model 1 and the learning target model has role information of the welding process for car model 1, the pre-learning model and the learning target model Although the feature set information of , it may be determined that the role information is different. In this case, the processor 110 may determine the pre-learning model having the role information of the car model 1 having the higher-level role information as the all-data learning model, and determine the all-data learning model as the seed model for transfer learning. can

Alternatively, the processor 110 may determine the pre-learning model as the seed model when the feature set information of the pre-learning model and the learning target model do not match and the role information of the pre-learning model and the learning target model are similar. . 4, when the pre-learning model has role information of the painting process for car model 1 and the learning target model has role information of the painting process for car model 2, the processor 110 is Although the feature set information of the learning model and the model of the learning target do not match, it may be determined that the role information is similar. In this case, the processor 110 may determine a pre-learning model having role information of the painting process for the vehicle model 1 as a seed model for performing transfer learning.

In the above description, the matching of the feature set information is exemplified when the process object information 320 is the same on the role information, and the similar role information is exemplified as the case where the process information 330 is the same. The standards regarding the similarity of the standards and role information are not limited thereto. That is, the feature sets may be matched when the two models have the same machine device information 310 according to the types of input data or input vectors of the pre-learning model and the learning target model. Alternatively, the feature sets can be matched only when the two models have the same process information 330 according to the types of input data or input vectors of the pre-learning model and the learning target model. In addition, in the above-described example, the feature set is 'matched' as a case in which the feature sets are the same, but this is only an example of a criterion regarding whether the feature sets match, and the criterion regarding whether the feature sets are matched is not limited thereto. will say no

In addition, in the above description, the similarity between the role information of the pre-learning model and the learning target model is exemplified according to whether the process information 330 matches, but the criterion for the processor 110 to determine whether the role information is similar is It should not be limited to this.

When transfer learning is performed on the learning target model only with the matching of feature sets, the purpose (output data) is different between the learning target model and the pre-learning model, and the pre-learning model contains the There is a problem that knowledge is not taken into account. Therefore, in order to overcome this limitation, it is possible to determine the seed model by considering not only the matching degree of feature set information but also the role information of the pre-learning model and the learning target model. In this case, since the learning goal of the learning target model and the goal of the pre-learning model are similar, the knowledge contained in the pre-learning model is more likely to coincide with the knowledge about the feature extraction that the learning target model needs to learn. Therefore, transfer learning for the learning target model can be performed more effectively.

5 is a schematic diagram illustrating an initial state of a network in which transfer learning is performed according to an embodiment of the present disclosure.

In the present disclosure, the processor 110 may determine an initial state of transfer learning for performing transfer learning on the learning target model. In this case, the processor 110 may use some or all of the weights of the seed model to determine the initial state of transfer learning of the learning target model. Here, using a portion of the weight of the seed model may mean using weight information of a network connected to a portion of the common feature set 230 . Alternatively, using a part of the weight of the seed model may mean using weight information on some of the entire layers of the network of the seed model related to the common feature set 230 . Here, using the weight information may mean determining the weight information of the seed model as the initial weight information of the learning target model. In the present disclosure, the first transfer learning initial state 400 includes only the weight information in the first layer (that is, weight information from the input layer to the first hidden layer) among the weight information of the seed model, the initial weight information of the learning target model. It may mean the state of the learning target model of the state determined as . In one embodiment, performing transfer learning by applying the weight information of the seed model to the learning target model includes: from the feature set information of the seed model and the feature set information of the learning target model, the seed model and the learning target model It may include extracting a common feature set and applying weight information of some layers or all layers related to the common feature set of the seed model as initial weight information of the learning target model. Here, applying the weight information of some layers related to the common feature set of the seed model as the initial weight information of the learning target model is the weight information in the input layer related to the common feature set of the seed model (that is, the first hidden in the input layer). weight information to the layer) may be applied as initial weight information of the learning target model.

As described above, when the initial layers of the seed model are determined as initial weight information of the learning target model, only the knowledge of the seed model regarding abstract feature extraction from input data may be transferred to the learning target model. Therefore, when the feature set information and role information of the learning target model and the seed model are reviewed, knowledge about specific feature extraction and judgment is required to learn by itself, only the weight information of the initial layers of the seed model or the input layer is learned After determining the initial weight information of the target model, transfer learning can proceed.

Furthermore, the amount of weight information of the seed model (or the number of layers of the seed model) to be applied as the initial weight information of the learning target model may vary according to the ratio of the common feature set. That is, the processor 110 may determine the size of the common feature set (ie, the number of common features) and determine the number of layers of the seed model to be applied as initial weight information of the learning target model based on this. For example, the processor 110 may make the number of features included in the common feature set 230 proportional to the number of weight layers applied from the seed model to the learning target model.

As the size of the common feature set 230 between the seed model and the learning target model increases, transfer learning is more effective in transferring the knowledge of the seed model not only for feature extraction at an abstract level but also for specific feature extraction and judgment to the learning target model can make it possible to perform Accordingly, transfer learning can be performed more effectively by adjusting the amount of weight information to be transferred from the seed model according to the size of the common feature set 230 .

6 is a schematic diagram illustrating an initial state of a network in which transfer learning is performed according to an embodiment of the present disclosure.

As described above, in the present disclosure, the processor 110 may determine an initial state of transfer learning for performing transfer learning on the learning target model. In this case, the processor 110 may use some or all of the weights of the seed model to determine the initial state of transfer learning of the learning target model. Here, using all the weights of the seed model may mean using weight information of networks connected to all of the common feature sets 230 . Alternatively, using all weights of the seed model may mean using weight information for all layers of the seed model network related to the common feature set 230 . Here, using the weight information may mean determining the weight information of the seed model as the initial weight information of the learning target model. In the present disclosure, the second transfer learning initial state 500 may refer to the state of the learning object model in a state in which all weight information of the seed model is determined as the initial weight information of the learning object model. In one embodiment, performing transfer learning by applying the weight information of the seed model to the learning target model includes: from the feature set information of the seed model and the feature set information of the learning target model, the common of the seed model and the learning target model This may include extracting the feature set and applying weight information of all layers related to the common feature set of the seed model as initial weight information of the learning target model.

When all the knowledge of the seed model regarding a common feature set is transferred to the learning target model, all knowledge contained in the seed model can be transferred to the learning target model, so there may be an advantage in that effective transfer learning is possible. Not only this, but rather than simply importing all the structures of the pre-trained model, only the features common to the seed model and related networks (or weight information) are imported after separately building the learning target model. Unnecessary knowledge may not be transferred from this seed model. Therefore, more effective transfer learning can be performed compared to the existing transfer learning using the previously learned model as it is.

7 is a flowchart illustrating a process in which a processor performs transfer learning according to an embodiment of the present disclosure.

The processor 110 may analyze meta-information of one or more pre-learning models and meta-information of a learning target model (S100).

In the present disclosure, the pre-learning model and the learning target model may be constructed using an artificial neural network. In particular, the pre-learning model and the learning target model may be an artificial neural network model trained/trained to perform anomaly detection. The pre-learning model and the learning target model may include a feature set 200 related to each input vector, feature set information representing the feature set, and role information about a role performed by each model. In some embodiments of the present disclosure, the pre-learning model may be a pre-trained model that may be a seed model in transfer learning, and the learning target model may be a model on which transfer learning is to be performed. In some embodiments of the present disclosure, the feature sets of the prior learning model and the learning target model may match, match, or have some common feature sets. The processor 110 may determine the degree of similarity between the pre-learning model and the feature set of the learning target model by comparing each of the features of the pre-learning model and the learning target model or comparing feature set information. In addition, the pre-learning model and the learning target model include role information, and the processor 110 includes hierarchical information 300 , mechanical device information 310 , process object information 320 , and process information 330 included in the role information. can be used to determine whether role information is similar between the pre-learning model and the learning target model. The processor 110 according to some embodiments of the present disclosure may select a seed model to be used for transfer learning from pre-learning models by using meta information including the role information and feature set information.

In the present disclosure, meta information may be matched to a pre-learning model and a learning target model. Such meta information includes information about the types of input data and output data of the pre-training model and the learning target model, the learning method of the pre-learning model and the training target model, and the process and process in which the pre-training model and the training target model are used. It may include information about the object, the pre-learning model, and the machine device to which the learning target model is related. In particular, in the present disclosure, meta information may include feature set information and role information of the pre-learning model and the learning target model. The feature set information and role information will be described later in detail with reference to FIGS. 3 and 4 .

In the present disclosure, the feature set 200 may be input data or input vectors of the pre-training model and the learning target model. For example, the feature set 200 may be input data or input vectors for a prior learning model, a learning target model, or an all-data learning model. Such a feature set 200 may relate to an entire input vector. Also, this feature set 200 may be composed of one or more sub-feature sets. In this case, each of the sub-feature sets may be associated with a part of the input vector. For example, when the feature set 200 is a set of sub-input data recognized through a plurality of sensors, each of the sub-feature sets may correspond to any one of the plurality of sensors. That is, the features included in the sub-feature set may represent data sensed by any one of the plurality of sensors. For example, it is assumed that an image is taken of the side of the car in order to check the work result of the side of the car, and that the photographing is performed by three sensors. In this case, the feature set 200 may include all of the sub-feature sets recognized by each sensor. In this case, each sub-feature set may correspond to partial photographing data of the side of the vehicle recognized by each sensor.

Furthermore, the feature set information according to the present disclosure may be information on the form of input data or input vectors of one or more pre-learning models and each of the learning target models. That is, the feature set information may include information on the size of the feature set that is the input vector, the type of the feature set that is the input vector (image or voice, etc.). That is, the feature set information may be referred to as meta information about the input vector included in the corresponding feature set. Accordingly, when a plurality of feature sets exist, the processor 110 according to the present disclosure may determine the similarity of the feature sets between one or more pre-trained pre-learning models and the learning target model based on the feature set information. Based on this, the processor 110 may determine which pre-learning model to use to perform transfer learning on the learning target model. Furthermore, based on the similarity of the feature sets, the processor 110 may determine the number of layers of the pre-learning model from which weight information is to be obtained during transfer learning. Details on this will be described later.

In the present disclosure, role information may be information about a process, a machine device, and a task performed by the process or the machine device related to the pre-learning model or the learning target model. For example, assuming that the pre-learning model or the learning target model according to the present disclosure is an artificial neural network model that determines whether the painting result for the car model 1 is anonymous, the role information of the pre-learning model or the learning target model is It may be "painting process for car model 1". In the present disclosure, such a pre-learning model or a learning target model may be separately trained according to the types of input data and output data of the corresponding model and according to the hierarchy of role information. For example, assuming that the pre-training model or the learning target model is an artificial neural network model that determines whether the task result of the car model 1 is annomalous, the role information of the pre-training model or the learning target model is "car model 1 "It could be Alternatively, assuming that the pre-learning model or the learning target model is an artificial neural network model that determines whether the task result of the robot arm model A is annomalous, the role information of the pre-training model or the learning target model is "robot arm model A "It could be

As described above, each pre-learning model or learning target model may be trained using different data sets according to the hierarchy of role information. For example, when the role information is "painting process of car model 1", a pre-learning model or a learning target model corresponding to the role information may be trained using result data of the painting process of car model 1. Alternatively, the pre-learning model or the learning target model in the case where the role information is “car model 1” may be trained using both the results of the painting process of the car model 1 and the welding process of the car model 1. In addition, when the role information is “robot arm model A”, a pre-training model or a learning target model corresponding thereto may be trained using both the vehicle model 1 and the result data of the painting process and the welding process of the vehicle model 1.

The processor 110 may determine a seed model among one or more pre-learning models based on the analysis result of the meta information ( S200 ).

In the present disclosure, the seed model may be an artificial neural network model that transmits weight information to the learning target model in order to perform transfer learning. The processor 110 may determine a seed model for an arbitrary learning target model from one or more pre-learning models based on meta information, particularly, feature set information and role information included in the meta information. In an embodiment, the processor 110 may determine a pre-learning model that matches the learning target model and feature set information and has similar role information as the seed model. In another embodiment, when a pre-learning model that matches the learning target model and the feature set information and has different role information exists, the processor 110 uses the all-data learning model located in the upper hierarchy of the pre-learning model as a seed for transfer learning. can be determined by the model. In another embodiment, the processor 110 may determine a pre-learning model in which the learning target model and the feature set information match by a certain ratio or more (satisfying a preset criterion) and the role information is similar as the seed model for transfer learning. have.

The processor 110 may perform transfer learning on the learning target model using weight information of the seed model (S300).

8 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 generally as being 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 can be applied to single-processor or multiprocessor computer systems, minicomputers, mainframe computers as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, and the like. It will be appreciated that each of these may be implemented in other computer system configurations, including operable in conjunction 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 so as 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 further be 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 for external use within 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.

Those 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 For this purpose, 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 (11)

A computer program stored on a computer-readable storage medium, the computer program comprising instructions for causing one or more processors to perform transfer learning on a neural network, the instructions comprising:
Analyzing meta-information of one or more pre-learning models and meta-information of a learning target model - The meta information includes feature set information and role information of each of the one or more pre-learning models, and feature set information and roles of the learning target model contains information-;
determining a feature set similarity of the one or more prior learning models and the learning object model based on a common feature set of the one or more prior learning models and the learning object model;
comparing the role information of the one or more prior learning models and the role information of the learning target model;
determining a seed model among the one or more pre-trained models based on the determined feature set similarity and a result of the comparison; and
performing transfer learning on the learning target model using weight information of a layer related to the common feature set included in the seed model;
containing,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The feature set information is
Information about the shape of the input vector of each of the one or more pre-learning models and the learning target model,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The role information is
At least one of the pre-learning model or the learning target model includes at least one of a related process, a machine device, or a process object, and hierarchical information between the process, the machine device, and the process object,
A computer program stored in a computer-readable storage medium.
The method of claim 1,
The step of determining a seed model among the one or more pre-learning models comprises:
identifying a first pre-learning model in which feature set information satisfies a preset criterion; and
determining the seed model to be used for transfer learning based on a comparison result of the role information of the first pre-learning model and the learning target model;
containing,
A computer program stored in a computer-readable storage medium.
5. The method of claim 4,
The step of determining the seed model to be used for transfer learning based on a comparison result of the role information of the first pre-learning model and the learning target model includes:
determining an all-data learning model as the seed model when the feature set information of the first pre-learning model and the learning target model match, and the role information of the first pre-learning model and the learning target model are different ; or
When the feature set information of the first pre-training model and the learning target model do not match and the role information of the first pre-training model and the learning target model are similar, the first pre-learning model is used as the seed model determining;
containing,
A computer program stored in a computer-readable storage medium.
delete The method of claim 1,
The step of performing transfer learning on the learning target model using weight information of a layer related to the common feature set includes:
extracting a common feature set of the seed model and the learning target model from the feature set information of the seed model and the feature set information of the learning target model; and
applying weight information of some layers or all layers related to the common feature set of the seed model as initial weight information of the learning target model;
containing,
A computer program stored in a computer-readable storage medium.
8. The method of claim 7,
The step of applying weight information of some layers or all layers related to the common feature set of the seed model as initial weight information of the learning target model includes:
determining a size of a common feature set between the seed model and the learning target model based on the feature set information of the seed model and the feature set information of the learning target model;
determining the number of layers of the seed model to be applied as initial weight information of the learning target model in proportion to the size of the common feature set;
containing,
A computer program stored in a computer-readable storage medium.
A method for performing transfer learning on a neural network, performed by a computing device comprising at least one processor, the method comprising:
Analyzing meta-information of one or more pre-learning models and meta-information of a learning target model - The meta information includes feature set information and role information of each of the one or more pre-learning models, and feature set information and roles of the learning target model contains information-;
determining a feature set similarity of the one or more prior learning models and the learning object model based on a common feature set of the one or more prior learning models and the learning object model;
comparing the role information of the one or more prior learning models and the role information of the learning target model;
determining a seed model among the one or more pre-trained models based on the determined feature set similarity and a result of the comparison; and
performing transfer learning on the learning target model using weight information of a layer related to the common feature set included in the seed model;
containing,
A method for performing transfer learning on neural networks.
A computing device for performing transfer learning on a neural network, the computing device comprising:
Memory; and
processor;
including,
The processor is
Analyze meta information of one or more pre-learning models and meta-information of a learning target model, wherein the meta information includes feature set information and role information of each of the one or more pre-learning models, and feature set information and role information of the learning target model including-,
Based on a common feature set of the one or more prior learning models and the learning target model, determining the feature set similarity of the one or more prior learning models and the learning target model,
comparing the role information of the one or more pre-learning models and the role information of the learning target model,
determining a seed model among the one or more pre-trained models based on the determined feature set similarity and a result of the comparison, and
performing transfer learning on the learning target model using weight information of a layer related to the common feature set included in the seed model;
A computing device that performs transfer learning on neural networks.
A computer-readable recording medium storing a data structure corresponding to a parameter of a neural network that is at least partially updated in a learning process performed by a computing device, wherein the operation of the neural network is based at least in part on the parameter, the learning process silver,
reviewing one or more pre-learning models and feature set information and role information of the learning target model;
determining a feature set similarity of the one or more prior learning models and the learning object model based on a common feature set of the one or more prior learning models and the learning object model;
comparing the role information of the one or more prior learning models and the role information of the learning target model;
determining a seed model among the one or more pre-trained models based on the determined feature set similarity and a result of the comparison; and
performing transfer learning on the learning target model using weight information of a layer related to the common feature set included in the seed model;
containing,
A computer-readable recording medium having a data structure stored thereon.

Images