WO2024063096A1 - 情報処理システム、情報処理方法及びプログラム - Google Patents

情報処理システム、情報処理方法及びプログラム Download PDF

Info

Publication number
WO2024063096A1
WO2024063096A1 PCT/JP2023/034117 JP2023034117W WO2024063096A1 WO 2024063096 A1 WO2024063096 A1 WO 2024063096A1 JP 2023034117 W JP2023034117 W JP 2023034117W WO 2024063096 A1 WO2024063096 A1 WO 2024063096A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
information processing
predetermined
processing device
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/JP2023/034117
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
昌道 中村
雅一 西本
博隆 井上
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Morgenrot Inc
Original Assignee
Morgenrot Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Morgenrot Inc filed Critical Morgenrot Inc
Priority to JP2024548285A priority Critical patent/JPWO2024063096A1/ja
Priority to EP23868217.3A priority patent/EP4589491A4/en
Publication of WO2024063096A1 publication Critical patent/WO2024063096A1/ja
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/098—Distributed learning, e.g. federated learning

Definitions

  • the present invention relates to an information processing system, an information processing method, and a program.
  • Patent Document 1 describes a technique for generating a trained model for an accounting processing service using an accounting processing device on the cloud and for making a determination using the trained model.
  • the present invention was made in view of this situation, and aims to improve the convenience of system processing while preventing information leakage.
  • an information processing system includes: An information processing system including a first information processing device that executes a part of a predetermined process, and a second information processing device that executes the remaining part of the predetermined process,
  • the first information processing device includes: a first execution unit that executes a part of the predetermined process using predetermined information as input information, and outputs the predetermined information as output information indicating the execution result in a form that cannot be reproduced with only the output information; Equipped with
  • the second information processing device includes: acquisition means for acquiring the output information from the first information processing device; a second execution means for executing the remaining part of the predetermined process using the output information as input information; Equipped with
  • the information processing method and program according to one embodiment of the present invention correspond to the information processing apparatus according to one embodiment of the present invention described above.
  • the present invention makes it possible to prevent information leaks while improving the convenience of processing on the system.
  • FIG. 1 is a diagram showing an example of a system configuration of one embodiment of an information processing system according to the present invention.
  • 2 is a diagram illustrating an example of the hardware configuration of a server in the information processing system of FIG. 1.
  • FIG. 3 is a functional block diagram illustrating an example of a functional configuration of an information processing system including a server having the hardware configuration shown in FIG. 2.
  • FIG. 4 is a flowchart showing the operation of the information processing system having the hardware configuration of FIG. 2 and the functional configuration of FIG. 3.
  • FIG. 4 is a diagram showing a predetermined process in an information processing system having the hardware configuration of FIG. 2 and the functional configuration of FIG. 3.
  • FIG. 6 is a diagram illustrating how the predetermined process in FIG. 5 is divided into user terminal-side processing and server-side processing and executed.
  • FIG. 5 is illustrating an example of a system configuration of one embodiment of an information processing system according to the present invention.
  • FIG. 4 is a diagram illustrating an example of a learning process of a neural network in the information processing system of FIG. 3 .
  • 4 is a diagram illustrating an example of a process for extracting feature amounts from original data in the information processing system of FIG. 3.
  • FIG. 9 is a diagram illustrating an example of a process of converting a plurality of feature amounts extracted from different cross sections in the process of FIG. 8 into a matrix.
  • FIG. 7 is another example of the process shown in FIG. 6, and is a diagram illustrating an example of a process for converting, for example, a flow field of airflow into a feature amount other than shape data.
  • FIG. 2 is a diagram illustrating an embodiment in which feature data is extracted using a neural network of many layers in an information processing system according to the present invention.
  • FIG. 3 is a diagram illustrating an embodiment in which a learning phase of a neural network is performed by three networks in an information processing system according to the present invention.
  • FIG. 2 is a diagram showing an embodiment in which each of a plurality of layers of a learning section and the learning section itself are distributed and arranged in different servers in an information processing system according to the present invention.
  • FIG. 7 is a diagram showing an embodiment in which processing is divided into parts different from the example of FIG. 6 in the information processing system according to the present invention.
  • FIG. 1 is a diagram showing an example of the system configuration of an information processing system according to one embodiment of the present invention.
  • the information processing system in FIG. 1 is configured to include a server 1 and a user terminal 2.
  • the server 1 and the user terminal 2 are connected to be able to communicate with each other via a predetermined network N such as the Internet.
  • a predetermined network N such as the Internet.
  • the form of the network N is not particularly limited, and for example, Bluetooth (registered trademark), Wi-Fi, LAN (Local Area Network), the Internet, etc. can be adopted.
  • the information processing system analyzes the airflow of three-dimensional airfoil shape data through communication via the network N between the server 1 and the user terminal 2, and simulates (simulates) the flow field of the airflow. Executes predetermined processing such as processing.
  • the server 1 is an information processing device such as a smartphone, a tablet terminal, or a personal computer (PC) that is managed by a service provider, and has an application for the server provided by the service provider installed therein.
  • the server 1 executes the remaining processes among the predetermined processes in response to a request from the user. Specifically, while communicating with the user terminal 2 as appropriate, the server 1 generates, for example, a cross-sectional curve of a three-dimensional airfoil pseudo-shape from the feature data received from the user terminal 2 as the remaining processing.
  • Various control processes are executed, including a process of generating a performance parameter indicating the magnitude of airflow resistance with respect to the cross-sectional shape, or simulating (simulating) the airflow field according to the performance parameter.
  • the user terminal 2 is, for example, a terminal that accepts operations by a user, and is, for example, an information processing device such as a smartphone, a tablet terminal, or a personal computer (PC).
  • a user application provided by a service provider is installed on the user terminal 2.
  • the user terminal 2 executes some of the predetermined processes. Specifically, the user terminal 2 executes, for example, a process of extracting feature data from shape data stored in the user terminal 2 as part of the predetermined process while communicating with the server 1 as appropriate. do.
  • the user terminal 2 and the server 1 install a dedicated application provided by a service provider, and perform the first
  • the method was adopted as a "collaboration" method.
  • a second method may be adopted in which the user terminal 2 accesses a website disclosed by the server 1 and displays the website.
  • the user terminal 2 executes some of the predetermined processes and has the server 1 execute the remaining processes while communicating with the server 1 as appropriate.
  • the method by which the user terminal 2 and the server 1 cooperate to realize this service in this way is not limited to the first method.
  • a method in which the first method and the second method are appropriately combined can also be employed as the "collaboration" method.
  • the server 1 is the main body of processing, but this is merely an example, and it goes without saying that the user terminal 2 or other information processing device (server 1, etc.) may be used as appropriate during implementation.
  • server 1 in the information processing system in Figure 1 and explain the hardware configuration of server 1 below.
  • FIG. 2 is a diagram showing an example of the hardware configuration of a server in the information processing system shown in FIG.
  • the server 1 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a bus 14, an input/output interface 15, an output section 16, and an input section 1. 7 and , a storage section 18, a communication section 19, and a drive 20.
  • CPU Central Processing Unit
  • ROM Read Only Memory
  • RAM Random Access Memory
  • the CPU 11 executes various processes according to programs recorded in the ROM 12 or programs loaded into the RAM 13 from the storage section 18 .
  • the RAM 13 also appropriately stores data necessary for the CPU 11 to execute various processes.
  • the CPU 11, ROM 12, and RAM 13 are interconnected via a bus 14.
  • An input/output interface 15 is also connected to this bus 14 .
  • An output section 16 , an input section 17 , a storage section 18 , a communication section 19 , and a drive 20 are connected to the input/output interface 15 .
  • the output unit 16 is composed of a display such as a liquid crystal display device, a printer, a speaker, etc., and outputs various information.
  • the input unit 17 is constituted by input devices such as a keyboard and a mouse, and various information is inputted into the input unit 17 .
  • the storage unit 18 is configured with a dynamic random access memory (DRAM) or the like, and stores various data.
  • the communication unit 19 communicates with other devices via a network N including the Internet.
  • a removable medium 30 made of a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is appropriately installed in the drive 20.
  • the program read from the removable medium 30 by the drive 20 is installed in the storage unit 18 as necessary. Further, the removable medium 30 can also store various data stored in the storage section 18 in the same manner as the storage section 18.
  • FIG. 3 is a functional block diagram showing an example of the functional configuration of an information processing system including a server having the hardware configuration shown in FIG. 2.
  • the hardware elements of the server 1 are given k and the hardware elements of the user terminal 2 are given s. shall be explained as follows.
  • the storage unit 18s of the user terminal 2 stores a learning model 81 in addition to the original data to be processed. Furthermore, in the CPU 11s of the user terminal 2, the encoding unit 41 functions when executing predetermined processing.
  • the encoding section 41 includes an encoder 51 and a learning section 52.
  • the learning section 52 may be integrated into the learning section 72 on the server side, or vice versa, instead of being placed on the user terminal 2 side, depending on how the predetermined processing is separated.
  • the encoding unit 41 executes part of a specified process using data to be processed (original data), such as three-dimensional airfoil shape data, as input information, and outputs as output information indicating the results of the execution in a form in which the specified information cannot be reproduced by the output information alone.
  • Original data data to be processed
  • the specified processing is, for example, processing for analyzing the airflow of the three-dimensional airfoil shape data and simulating the behavior of the airflow
  • the processing performed on the user terminal 2 side is part of the specified processing, for example, processing for converting the original data into feature data in which the coordinate values of multiple points required to generate the cross-sectional curve of the three-dimensional airfoil are arranged as features.
  • "impossible to reproduce” means that it is impossible to reproduce within a realistic time frame and in a state where no information other than the output information is given.
  • the encoder 51 reads out the original data stored in the storage unit 18s, and generates digital data of “0” and “1” necessary for generating the curve of the cross section of the three-dimensional airfoil from the original data, Execute the process to generate the digital data, convert the "0" and "1” digital data into feature data (feature matrix), which is a feature in which the coordinate values of multiple points are arranged, and output. (See Figure 9).
  • the learning unit 52 causes the learning model 81 to learn in the learning phase. Furthermore, in the operation phase, the learning unit 52 inputs the feature data (feature matrix) generated by the encoder 51 to the learning model 81, and causes the learning model 81 to output feature data reduced in dimension. Specifically, the learning unit 52 uses the feature data (feature matrix) generated by the encoder 51 as input information to construct a neural network (see FIG. 7) consisting of three layers, for example, the first layer to the third layer. Weighting processing is performed on the first layer, and the execution results (feature amount data) are output.
  • a neural network see FIG. 7
  • the output information from the user terminal 2 is feature data, and by providing setting information and reproduction means (dedicated decoder, etc.) for reproduction on the server 1 side in advance, it is possible to reproduce the original data or data approximating it. It can be said to be a form in which the original data cannot be reproduced using feature data alone.
  • the output information (for example, feature data, etc.) output from the user terminal 2 is based on the target indicated by the predetermined information (for example, three-dimensional airfoil shape data, data on airflow analysis results, etc.) when the original data is used as predetermined information. ) can be said to be one or more feature data (features).
  • the learning unit 52 is arranged after the encoder 51, and the output information is the learning result during the predetermined machine learning (first layer learning only), but other information may be used.
  • the learning section 52 may output the results of learning the first layer and the second layer.
  • the learning section 52 may be deleted and the output of the encoder 51 itself may be output as output information (see FIG. 14).
  • the decoding unit 61 functions when executing predetermined processing.
  • an information acquisition section 71 In the decoding section 61, an information acquisition section 71, a learning section 72, a decoder 73, and a simulating section 74 function.
  • the decoding unit 61 acquires output information transmitted from the user terminal 2 via the Internet, and executes the remaining part of the predetermined process using the output information as input information.
  • the remaining part of the predetermined processing involves, for example, generating a curve of the cross-section DM (see FIG. 8) of the pseudo-shape of the three-dimensional airfoil D1 from the feature data and calculating performance parameters indicating the magnitude of airflow resistance with respect to the cross-sectional shape. It includes processing to generate or simulate the flow field of airflow according to the cross-sectional shape.
  • the information acquisition unit 71 acquires output information from the user terminal 2. Specifically, the information acquisition unit 71 acquires the feature amount data transmitted from the user terminal 2, and temporarily stores it in the storage unit 18k.
  • the learning unit 72 performs predetermined machine learning using the input feature data (teacher data) and outputs a learning result.
  • the learning unit 72 has a second layer and a third layer among the three layers, and, for example, processes the input feature data in the order of the layers (processing such as weighting and addition).
  • processing such as weighting and addition
  • feature amount data in a state where it is impossible to reproduce the same data as the original data existing on the user terminal 2 side is generated. That is, the predetermined machine learning is learning using N-layer (N is an integer value of 2 or more) learning models 81 and 91 (neural networks), and the output information from the user terminal 2 is the learning model of the neural network.
  • the remaining part of the predetermined processing is, for example, processing to obtain performance parameters by analyzing the flow of airflow applied from a predetermined direction to the shape data of the three-dimensional airfoil, etc., and the training data is input to the learning model 91 for learning. processing (processing to execute predetermined machine learning).
  • the decoder 73 reproduces (generates) pseudo-shape data of the three-dimensional airfoil from the learning results of the learning unit 72, analyzes the flow of airflow applied to the three-dimensional airfoil from a predetermined direction K, and obtains three-dimensional information about the flow of the airflow. Execute the process to obtain the performance parameters of the dimensional airfoil.
  • the simulator 74 simulates the flow of airflow applied to the three-dimensional airfoil from a predetermined direction. Specifically, the simulator 74 simulates a flow field NB in which airflow is applied from a predetermined direction K to the three-dimensional airfoil pseudo shape data generated by the decoder 73 (see FIG. 10). That is, the predetermined process includes a process of performing a simulation (for example, a process of analyzing the flow of airflow that hits a three-dimensional airfoil from a predetermined direction and simulating the behavior (flow field) of the airflow).
  • the user terminal 2 when executing a predetermined process (for example, a process of performing a performance test of a three-dimensional airfoil on data), the user terminal 2 Some of the processing is performed from the shape data (such as processing to generate feature data that is converted into feature data that arranges the coordinate values of multiple points necessary to generate the cross-sectional curve of a three-dimensional airfoil). , transmits (outputs) feature amount data indicating the execution result to the server 1.
  • Feature data is a form of data that cannot reproduce the original data by itself, and even if this feature data is leaked or stolen, the original data cannot be restored using feature data alone. .
  • the server 1 includes a decoding unit 61 and executes the remaining part of the predetermined processing using the feature amount data transmitted from the user terminal 2, so that the processing burden on the user side can be reduced. .
  • a decoding unit 61 executes the remaining part of the predetermined processing using the feature amount data transmitted from the user terminal 2, so that the processing burden on the user side can be reduced. .
  • FIG. 4 is a flowchart showing the operation of the information processing system having the functional configuration of FIG.
  • step S11 when executing a predetermined process (for example, a process of performing a three-dimensional airfoil performance test on data), in step S11, the encoding unit 41 extracts a part of the process from the three-dimensional airfoil shape data. and transmits (outputs) feature amount data indicating the execution results to the server 1.
  • a predetermined process for example, a process of performing a three-dimensional airfoil performance test on data
  • step S12 when the decoding unit 61 acquires the feature data transmitted from the user terminal 2, in step S13, the feature data is used as input information to perform the remaining part of the predetermined process. and transmits (outputs) the execution result of the predetermined process to the user terminal 2.
  • the execution result of the predetermined process is, for example, a performance parameter obtained as a result of the process executed by the decoder 73 or a simulation result (simulation result) of an airflow field obtained as a result of the process executed by the simulator 74.
  • the user terminal 2 Upon receiving the execution result of the predetermined process transmitted from the server 1, the user terminal 2 displays the execution result on the screen in step S14.
  • this information processing system by executing part of the processing on the user terminal 2 side and the remaining part of the processing on the server 1 side, the original data held by the user terminal 2 can be By having the user terminal 2 cause the server 1 to execute processing while preventing leakage, the processing burden on the user can be reduced.
  • FIG. 5 is a diagram for explaining an overview of predetermined processing when a three-dimensional airfoil is assumed as an example of a machine shape in an information processing system having the hardware configuration of FIG. 2 and the functional configuration of FIG. 3. .
  • the shape of the three-dimensional airfoil D1 is input to the encoder D3 as three-dimensional shape data (original data) and converted into digital data of "0" and "1".
  • original data When the original data is input to the learning model, it is converted into data such as voxel, mesh, and function expressions.
  • a three-dimensional airfoil D1 is divided into rectangular parallelepiped voxels D2.
  • the shape data divided as voxels D2 is input to the machine learning model through an encoder D3.
  • FIG. 5 an example of a learning model of an autoencoder including an encoder D3 and two layers D4 and D5 is shown.
  • a neural network or a deep neural network separate from the encoder D3 may be used, or another learning model such as a GAN may be used.
  • the output as a processing result may be an image of a flow field or a plurality of parameters instead of a performance parameter.
  • there may be no encoder or decoder and shape data is input to the layer part and output is obtained.
  • the shape data D2 read from the storage unit 18s is input to the layers D4 and D5 of the learning model through the decoder D3, and is reduced in dimension as a feature amount. This feature quantity is decoded by a decoder D6 and output as a performance parameter.
  • the output performance parameters are assumed to be fluid performance such as resistance, and structural performance such as strength. Furthermore, although the above-mentioned process has been shown as an example of mechanical design, it can be applied not only to this example but also, for example, to the fields of materials and drug discovery. In that case, parameters are set for the purpose of design and development in that field. Further, the predetermined information to be input may include not only shape data but also molecular structure, chemical formula, etc.
  • the learning model in Figure 5 is trained to output performance parameters for input shape data, but in the operational phase, this learning model is used to quickly estimate the performance of the design shape. . When using the learning models in the actual operation phase, all the learning models may be downloaded to local systems (for example, the user terminals 2 in FIGS. 1 and 3), and learning may be executed in each of the local systems.
  • FIG. 6 is a diagram illustrating how the predetermined process in FIG. 5 is divided into a process on the user terminal side (local system) and a process on the server side (cloud system) and executed.
  • a part of the predetermined processing in FIG. 5 is performed on the user terminal 2 side and a part of the remaining processing is performed on the server 1 side, for example, as shown in FIG. Design information
  • FIG. Design information is input to the encoder D3, inputted to the first layer D4 of the learning model, and processing until feature data is output from the first layer D4 of the learning model, that is, the shape data is input to the encoder D3 and the first layer D4 of the learning model.
  • the user terminal 2 side performs the processing up to conversion into feature amounts in layer D4.
  • the remaining part of the process is that the server 1 acquires the feature data transmitted from the user terminal 2, and in the server 1, the feature data is input to the second layer D5 of the learning model as input information.
  • the processing is divided (separated) between the first layer D4 and the second layer D5 among the plurality of layers of the learning section, but the part to be separated may be any part, for example, the encoder It may be divided between D3 and the first layer D4 of the learning section, or it may be divided between the second layer D5 and the decoder D6. Furthermore, the position at which the image is divided may change depending on the learning method. At least the steps after conversion into feature amounts may be processed on a cloud system such as the server 1.
  • FIG. 7 is a conceptual diagram of a basic learning model (neural network).
  • the neural network in this example is a three-layer neural network consisting of a first layer, a second layer, and a third layer. Each node is fully connected to the next layer's nodes.
  • data input to the neural network is assigned to nodes of the first layer.
  • the value is converted using the weighting function w and the bias b and is input to the second layer.
  • the converted value is uploaded from the user terminal 2 (local system) to the server 1 (cloud system), and the processing from the second layer onward is performed by the server. 1 is executed.
  • the values of the second layer are transformed by nonlinear transformation, weight function w', and bias b', and the results are output to the third layer.
  • the results are downloaded from the server 1 to the user terminal 2, and the weight function w and bias b are updated. Updated.
  • a learning model is generated, and this learning model is used on the user terminal 2 in the operation phase.
  • This method is just an example of using a neural network, and even when using a deep neural network, GAN, etc., the user terminal 2 and server 1 are switched during the processing of the learning model.
  • the analysis method also requires the use of a learning model that matches the dimensions of the features.
  • a learning model that matches the dimensions of the features.
  • FIG. 8 is a diagram for explaining the process of extracting feature points from the original data (three-dimensional airfoil shape data).
  • FIG. 9 is a diagram showing an example of a process for converting a plurality of feature amounts extracted from different cross sections in the process of FIG. 8 into a matrix.
  • FIG. 10 is a diagram illustrating an example of converting an airflow field into a feature amount.
  • an airflow such as wind is applied to a three-dimensional airfoil D1 from a predetermined direction K, and the flow field NB is converted into a feature quantity. do.
  • the flow field NB such as a vortex is generated.
  • Such a flow field NB can be decomposed into feature quantities using methods such as Proper Orthogonal Decomposition (POD), Dynamic Mode Decomposition (DMD), and Singular Value Decomposition (SVD). , and finally the matrix notation and It is possible to do so.
  • POD Proper Orthogonal Decomposition
  • DMD Dynamic Mode Decomposition
  • SVD Singular Value Decomposition
  • FIG. 11 is a diagram showing an embodiment in which feature data is extracted using a neural network of many layers in an information processing system according to the present invention.
  • a three-layer learning section consisting of a first layer D11, a second layer D12, and a third layer D13 is arranged in the user terminal 2, while the server 1
  • a two-layer learning section consisting of a fourth layer D14 and a fifth layer D15 is arranged.
  • a three-layer learning unit (neural network) is used to generate feature quantities in the user terminal 2, and the generated feature quantity data is transferred to a two-layer learning unit (for performance parameter learning) in the server 1. neural network).
  • shape data is input to a three-layer learning unit (neural network) that generates feature data, and the optimal feature data for learning in the server 1 is input. is learned to output.
  • the three-layer learning unit generates feature data using the learned learning model, and uploads the feature data to the server 1.
  • the server 1 learns the relationship between the feature data input to the two-layer learning unit and the performance parameters.
  • shape data is input as input information, and performance parameters are estimated from the input information through a combination process of the learning model generated on the user terminal 2 and the learning model generated on the server 1. be able to.
  • FIG. 12 is a diagram showing an embodiment in which the learning phase of the neural network is performed by three networks in the information processing system according to the present invention.
  • the information processing system of this embodiment arranges the first layer D21 of the neural network in the user terminal 2 connected to the first network, and connects it to a second network different from the first network.
  • the second layer D22 is placed in a cloud server that is connected to a third network
  • the third layer D23 is placed in a server 1 that is connected to a third network different from the second network.
  • FIG. 13 is a diagram showing an embodiment in which each of the plurality of layers of the learning section and the learning section itself are distributed and arranged in different servers in the information processing system according to the present invention.
  • a plurality of servers 1a and 1b are connected to a user terminal 2.
  • a first layer D31 and a second layer D32 of a learning section are arranged in the server 1a.
  • a third layer D33 and a fourth layer D34 of a learning section are arranged in the server 1b.
  • a learning process is executed on the uploaded feature data in each of the servers 1a and 1b, and the respective The performance parameters (output information) of the processing execution results are output from the servers 1a and 1b.
  • the learning process of the learning unit is illustrated, but the learning process is not limited to the learning process, and other processes may be executed.
  • each of the servers 1a and 1b may be caused to execute the same process and the results of each process may be compared.
  • the processing is separated into two servers 1a and 1b, but the processing is not limited to two servers 1a and 1b, and the processing may be separated into a plurality of servers.
  • FIG. 14 is a diagram showing an embodiment in which processing is divided into parts different from the example of FIG. 6 in the information processing system according to the present invention.
  • a part of the predetermined processing is performed on the user terminal 2 side, and a part of the remaining processing is performed on the server 1 side.
  • FIG. 6 a case has been described in which the processing is divided (separated) between the first layer D4 and the second layer D5 of the learning model, but other methods may be used. For example, as shown in FIG.
  • only the encoder D3 may be disposed in the user terminal 2, and the first layer D4, second layer D5, and decoder D6 of the learning section may be disposed in the server 1.
  • the encoder D3 in the user terminal 2, when the shape data (design information) to be processed is input to the encoder D3, the encoder D3 generates digital data with shape data of "0" and "1” and outputs it to the server 1. be done.
  • the digital data of "0" and "1" sent from the user terminal 2 is input to the learning section, and the remaining part of the process is executed.
  • the server 1 when digital data is input to the learning unit, processing is executed in the order of the first layer D4 and the second layer D5, and feature data is output from the second layer D5 to the decoder D6. be done.
  • the decoder D6 restores the pseudo shape from the input feature data, and outputs performance parameters and simulation results corresponding to the pseudo shape as output information.
  • the functional block diagram shown in FIG. 3 is merely an example and is not particularly limited. In other words, it is sufficient that the information processing system shown in FIG. 3 is equipped with a function that can execute the series of processes described above as a whole, and what kind of functional blocks and databases are used to realize this function is particularly dependent on FIG. 3.
  • the example is not limited to.
  • the locations of the functional blocks and the database are not limited to the example shown in FIG. 3, and may be arbitrary.
  • one functional block and database may be configured by a single piece of hardware, a single piece of software, or a combination thereof.
  • a program constituting the software is installed on a computer or the like from a network or a recording medium.
  • the computer may be a computer built into dedicated hardware. Further, the computer may be a computer that can execute various functions by installing various programs, such as a server, a general-purpose smartphone, or a personal computer.
  • Recording media containing such programs are not only comprised of removable media that is distributed separately from the device itself in order to provide the program to each user, but also are pre-installed in the device body and delivered to each user. Consists of provided recording media, etc.
  • the information processing apparatus to which the present invention is applied only needs to have the following configuration, and can take various embodiments.
  • the information processing system of one embodiment of the present invention includes: Some of the predetermined processing (e.g., obtaining performance parameters of a three-dimensional airfoil, analyzing the flow of airflow applied to a three-dimensional airfoil from a predetermined direction to simulate the behavior of the airflow, etc.)
  • a first information processing device for example, the user terminal 2 in FIG.
  • an information processing system including a second information processing device (for example, server 1 in FIG. 3) that executes The first information processing device (for example, the user terminal 2 in FIG.
  • predetermined information for example, processing target data such as shape data of a three-dimensional airfoil
  • a part of the predetermined processing for example, a curve of the cross section DM of the pseudo shape of the three-dimensional airfoil D1 (see FIG. 8)
  • processing to convert the coordinate values of multiple points required for generation into feature data is executed, and output information indicating the execution result (e.g. feature data etc.), a first execution unit (for example, the encoding unit 41 in FIG. 3) that outputs the predetermined information in a form that cannot be reproduced using only the output information; Equipped with The second information processing device (for example, the server 1 in FIG.
  • an acquisition unit for example, the information acquisition unit 71 in FIG. 3 that acquires the output information (for example, feature data, etc.) from the first information processing device (for example, the user terminal 2 in FIG. 3); Using the output information (for example, feature data, etc.) as input information, the remaining part of the predetermined processing (generating a curve of a cross-section of a pseudo-shaped three-dimensional airfoil from the feature data, and generating airflow resistance with respect to the cross-sectional shape) a second execution means (for example, the decoder 73 in FIG. 3), Equipped with In this case, the first information processing device (for example, the user terminal 2 in FIG.
  • predetermined information for example, shape data of a 3D model of a wing, etc.
  • predetermined processing for example, a cross section of a 3D model
  • output information indicating the execution results e.g. feature data, etc.
  • the second information processing device for example, the server 1 in FIG. 3
  • the second information processing device for example, the server 1 in FIG.
  • the output information for example, feature data, etc.
  • a process of generating a cross-sectional curve of a pseudo-shape, generating a performance parameter indicating the magnitude of airflow resistance with respect to the cross-sectional shape, and simulating a flow field according to the performance parameter is executed.
  • the first information processing device for example, the user terminal 2 in FIG. 3
  • the original specific shape cannot be restored, and if a third party were to Even if the feature data is leaked, there is no risk that the shape data itself stored in the first information processing device (for example, the user terminal 2 in FIG.
  • the second information processing device (for example, the user terminal 2 in FIG. 3) will be leaked, and the second information processing device (for example, the user terminal 2 in FIG. 3) The remaining part of the predetermined processing can be executed in the server 1, etc.). Furthermore, in the worst case scenario, even if the shape data is stolen, it is difficult for a third party to misuse the data unless the specifications (decoding rules, etc.) of the shape data are revealed. As a result, the first information processing device (for example, the user terminal 2 in FIG. 3) can prevent the leakage of predetermined information (processing target data) held by the first information processing device (for example, the user terminal 2 in FIG. 3). By having the second information processing device (for example, server 1 in FIG. 3) execute processing, the processing load on the first information processing device (for example, user terminal 2 in FIG. 3) can be reduced. The convenience of processing can be improved.
  • the output information (e.g., feature data) from the first information processing device is based on the target indicated by the predetermined information (e.g., three-dimensional airfoil shape data, airflow analysis results, etc.). Any one or more feature data (feature data, etc.) is sufficient.
  • the information processing device (for example, server 1 in FIG. 6) is The remaining part of the predetermined processing (for example, the processing to obtain performance parameters by analyzing the airflow applied to the three-dimensional airfoil from a predetermined direction) is the processing to perform predetermined machine learning (learning from training data). processing to input to the model 91 and learn it), It is sufficient that the output information (for example, feature data, etc.) from the first information processing device (for example, the user terminal 2 in FIG. 3) is a learning result during the predetermined machine learning.
  • predetermined processing for example, the processing to obtain performance parameters by analyzing the airflow applied to the three-dimensional airfoil from a predetermined direction
  • predetermined machine learning learning from training data.
  • the output information for example, feature data, etc.
  • the information processing device (for example, server 1 in FIG. 6) is The predetermined machine learning is learning using a neural network of N layers (N is an integer value of 2 or more), The output information from the first information processing device may be output information from K layers (K is an integer value of 1 or more and less than N) of the neural network.
  • the information processing device (for example, server 1 in FIG. 6) is
  • the predetermined process may include a process of performing a simulation (for example, a process of analyzing the flow of airflow that hits the three-dimensional airfoil from a predetermined direction and simulating the behavior (flow field) of the airflow).
  • An information processing method executed by the information processing system of one embodiment of the present invention is: A first step that executes a part of a predetermined process (for example, obtaining performance parameters of a three-dimensional airfoil, analyzing the flow of airflow applied to a three-dimensional airfoil from a predetermined direction, and simulating the behavior of the airflow, etc.) Information executed by an information processing system including an information processing device (for example, the user terminal 2 in FIG. 3) and a second information processing device (for example, the server 1 in FIG. 3) that executes the remaining part of the predetermined process.
  • the steps executed by the first information processing device for example, the user terminal 2 in FIG.
  • a feature in which coordinate values of a plurality of points necessary for generating a part of the predetermined processing (for example, a curve of a cross section of a 3D model) are arranged using predetermined information (for example, shape data of a 3D model of a wing, etc.) as input information.
  • a first execution step (for example, a process of converting the predetermined information into feature data converted into a quantity) and outputting the predetermined information in a form that cannot be reproduced with only the output information as output information indicating the execution result (e.g., 4 step S11 etc.), including;
  • the steps executed by the second information processing device include: an acquisition step (for example, step S12 in FIG.
  • a second execution step (for example, step S13 in FIG. 4), including, be able to.
  • the program of one embodiment of the present invention is A first step that executes a part of a predetermined process (for example, obtaining performance parameters of a three-dimensional airfoil, analyzing the flow of airflow applied to a three-dimensional airfoil from a predetermined direction, and simulating the behavior of the airflow, etc.) It is applied to an information processing system including an information processing device (for example, the user terminal 2 in FIG. 3) and a second information processing device (for example, the server 1 in FIG. 3) that executes the remaining part of the predetermined process.
  • a program, A computer for example, the CPU 11s of the user terminal 2 in FIG. 3) that controls the first information processing device (for example, the user terminal 2 in FIG.
  • a part of the predetermined processing using predetermined information for example, shape data of a three-dimensional airfoil, etc.
  • predetermined information for example, shape data of a three-dimensional airfoil, etc.
  • input information generating a pseudo-curve of the cross-section of the three-dimensional airfoil from the feature data and determining the magnitude of resistance with respect to the cross-sectional shape
  • a first execution step for example, step S11 in FIG. etc
  • Execute control processing including A computer (for example, the CPU 11k of the server 1 in FIG. 3) that controls the second information processing device (for example, the server 1 in FIG. 3), an acquisition step (for example, step S12 in FIG.
  • a second execution step for example, step S13 in FIG. 4 of performing a process of generating a performance parameter shown in FIG. execute control processing including be able to.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Image Analysis (AREA)
PCT/JP2023/034117 2022-09-20 2023-09-20 情報処理システム、情報処理方法及びプログラム Ceased WO2024063096A1 (ja)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2024548285A JPWO2024063096A1 (https=) 2022-09-20 2023-09-20
EP23868217.3A EP4589491A4 (en) 2022-09-20 2023-09-20 INFORMATION PROCESSING SYSTEM, INFORMATION PROCESS AND PROGRAM

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2022149423 2022-09-20
JP2022-149423 2022-09-20

Publications (1)

Publication Number Publication Date
WO2024063096A1 true WO2024063096A1 (ja) 2024-03-28

Family

ID=90454688

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2023/034117 Ceased WO2024063096A1 (ja) 2022-09-20 2023-09-20 情報処理システム、情報処理方法及びプログラム

Country Status (3)

Country Link
EP (1) EP4589491A4 (https=)
JP (1) JPWO2024063096A1 (https=)
WO (1) WO2024063096A1 (https=)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018011842A1 (ja) * 2016-07-11 2018-01-18 株式会社Ｕｅｉ 階層ネットワークを用いた演算処理システム
JP2018097813A (ja) 2016-12-16 2018-06-21 ファーストアカウンティング株式会社 会計処理装置、会計処理システム、会計処理方法、及び会計処理プログラム
JP2020069492A (ja) * 2018-10-30 2020-05-07 ファナック株式会社 加工条件設定装置及び三次元レーザ加工システム
JP2020155011A (ja) * 2019-03-22 2020-09-24 株式会社中電工 図面学習装置
JP2021135739A (ja) * 2020-02-27 2021-09-13 株式会社日立製作所 運転状態分類システム、および、運転状態分類方法
JP2022076477A (ja) * 2020-11-09 2022-05-19 キヤノンメディカルシステムズ株式会社 医用情報処理装置、医用情報処理システム及び医用情報処理方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018011842A1 (ja) * 2016-07-11 2018-01-18 株式会社Ｕｅｉ 階層ネットワークを用いた演算処理システム
JP2018097813A (ja) 2016-12-16 2018-06-21 ファーストアカウンティング株式会社 会計処理装置、会計処理システム、会計処理方法、及び会計処理プログラム
JP2020069492A (ja) * 2018-10-30 2020-05-07 ファナック株式会社 加工条件設定装置及び三次元レーザ加工システム
JP2020155011A (ja) * 2019-03-22 2020-09-24 株式会社中電工 図面学習装置
JP2021135739A (ja) * 2020-02-27 2021-09-13 株式会社日立製作所 運転状態分類システム、および、運転状態分類方法
JP2022076477A (ja) * 2020-11-09 2022-05-19 キヤノンメディカルシステムズ株式会社 医用情報処理装置、医用情報処理システム及び医用情報処理方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4589491A4

Also Published As

Publication number Publication date
EP4589491A4 (en) 2025-12-24
JPWO2024063096A1 (https=) 2024-03-28
EP4589491A1 (en) 2025-07-23

Similar Documents

Publication Publication Date Title
US11380034B2 (en) Semantically-consistent image style transfer
US10699388B2 (en) Digital image fill
KR102107709B1 (ko) 공간 트랜스포머 모듈들
KR20190117713A (ko) 신경망 아키텍처 최적화
EP3701429A1 (en) Auto-regressive neural network systems with a soft attention mechanism using support data patches
US11574019B2 (en) Prediction integration for data management platforms
WO2020237188A1 (en) Fully attentional computer vision
KR20190031318A (ko) 도메인 분리 뉴럴 네트워크들
KR20240154469A (ko) 확산 모델을 사용한 텍스트 기반 실제 이미지 편집
US20260037543A1 (en) Machine learning for optimized learning of human-understandable logical rules from medical or other data
CN113836425A (zh) 基于异构图注意力网络的电影推荐方法、装置及设备
US11087505B2 (en) Weighted color palette generation
CN117934254A (zh) 水印处理模型训练方法、水印处理方法、装置和设备
Agulhari et al. Robust real‐time parameter estimation for linear systems affected by external noises and uncertainties
CN119830033B (zh) 一种支持动态上下文的知识追踪方法及系统
WO2024063096A1 (ja) 情報処理システム、情報処理方法及びプログラム
Kefeng et al. CNN Based No‐Reference HDR Image Quality Assessment
CN116860304A (zh) 云服务更新方法、装置、计算机设备和存储介质
CN120803899A (zh) 测试用例生成方法、装置及电子设备
CN115187665B (zh) 点云数据处理方法、装置、电子设备及存储介质
CN117670878A (zh) 一种基于多模态数据融合的VOCs气体检测方法
CN114297402B (zh) 面向链接预测的交互式多模态融合方法及装置
Wang et al. Online mobile teaching methods based on Android in the 5G environment
CN118159984A (zh) 黑盒机器学习模型的新特征
CN117009205A (zh) 接口模拟方法、系统及计算机设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23868217

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2024548285

Country of ref document: JP

WWE Wipo information: entry into national phase

Ref document number: 2023868217

Country of ref document: EP

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2023868217

Country of ref document: EP

Effective date: 20250417

WWP Wipo information: published in national office

Ref document number: 2023868217

Country of ref document: EP