WO2019211909A1 - Dispositif d'apprentissage, dispositif de vérification, système et procédé de traitement de données - Google Patents

Dispositif d'apprentissage, dispositif de vérification, système et procédé de traitement de données Download PDF

Info

Publication number
WO2019211909A1
WO2019211909A1 PCT/JP2018/017538 JP2018017538W WO2019211909A1 WO 2019211909 A1 WO2019211909 A1 WO 2019211909A1 JP 2018017538 W JP2018017538 W JP 2018017538W WO 2019211909 A1 WO2019211909 A1 WO 2019211909A1
Authority
WO
WIPO (PCT)
Prior art keywords
dnn
verification
data
generation unit
neural network
Prior art date
Application number
PCT/JP2018/017538
Other languages
English (en)
Japanese (ja)
Inventor
直大 澁谷
杉本　和夫
彰 峯澤
守屋　芳美
Original Assignee
三菱電機株式会社
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 三菱電機株式会社 filed Critical 三菱電機株式会社
Priority to PCT/JP2018/017538 priority Critical patent/WO2019211909A1/fr
Priority to JP2020517000A priority patent/JP6742565B2/ja
Publication of WO2019211909A1 publication Critical patent/WO2019211909A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N99/00—Subject matter not provided for in other groups of this subclass

Definitions

  • the present invention relates to a learning device for learning a deep neural network (hereinafter referred to as DNN).
  • DNN deep neural network
  • IoT Internet of Things
  • a server receives and accumulates data acquired by a client via the Internet, and the server performs signal processing and statistical processing of the accumulated data in order to find some knowledge or meaning from the accumulated data.
  • the system is drawing attention.
  • a technique has been proposed in which a client converts data into metadata and then transmits the data to the server, thereby reducing the processing load and transmission load of the server and real-time processing.
  • a technique has been proposed in which a client converts data into metadata using DNN.
  • Patent Document 1 describes a system in which a server having a higher processing performance and memory capacity than a client learns DNN and transmits the learned DNN to the client for use.
  • This invention solves the said subject, and aims at obtaining the learning apparatus which can simplify the verification process of DNN.
  • the learning device includes a DNN generation unit, a verification data generation unit, and a data transmission unit.
  • the DNN generation unit generates a deep neural network learned to satisfy the requirements in response to a request received from the verification device that verifies whether the deep neural network satisfies the requirements using the verification data.
  • the verification data generation unit generates verification data used for verification of whether or not the deep neural network generated by the DNN generation unit satisfies the requirements.
  • the data transmission unit transmits the deep neural network generated by the DNN generation unit and the verification data generated by the verification data generation unit to the verification device.
  • the learning device generates verification data used for verification of whether or not the DNN satisfies the requirement and transmits the data to the verification device, so that the DNN verification process can be simplified.
  • FIG. 3 is a flowchart illustrating a data processing method according to the first embodiment. It is a flowchart which shows a DNN production
  • FIG. 12A is a block diagram illustrating a hardware configuration that implements the functions of the server according to the first embodiment.
  • FIG. 12B is a block diagram illustrating a hardware configuration for executing software that implements the functions of the server according to the first embodiment.
  • FIG. 12C is a block diagram illustrating a hardware configuration for realizing the functions of the client according to the first embodiment.
  • FIG. 12D is a block diagram illustrating a hardware configuration for executing software that implements the functions of the client according to the first embodiment.
  • FIG. 1 is a block diagram showing a configuration of a data processing system 1 according to Embodiment 1 of the present invention.
  • the data processing system 1 includes a server 2 and a client 3, and the DNN generated by the server 2 is used by the client 3.
  • the server 2 generates a DNN model and verification data in accordance with a command received from the user, and transmits the DNN model and verification data to the client 3.
  • the client 3 performs the performance verification of the DNN model using the verification data received from the server 2, and determines whether the DNN model can be operated in various applications based on the verification result.
  • the data processing system 1 is a client server model system in which the server 2 is a master and the client 3 is a slave.
  • the client 3 may include an instruction received from the user in the request information and transmit the request information to the server 2, and the server 2 may generate the DNN model and the verification data according to the instruction from the user included in the request information.
  • the client 3 is a master and the server 2 is a slave.
  • the client 3 is sometimes called an edge device.
  • the server 2 is a learning device that learns a DNN to be transmitted to the client 3, and includes a DNN generation unit 20, a verification data generation unit 21, a data transmission unit 22, and a DNN database 23.
  • the client 3 is a verification device that verifies whether the DNN received from the server 2 satisfies the requirement, and includes a data verification unit 30 and a request transmission unit 31.
  • the DNN generation unit 20 learns the DNN so as to satisfy the requirements included in the command from the user in response to the DNN generation request from the client 3.
  • the verification data generation unit 21 generates verification data used for verification of whether or not the DNN generated by the DNN generation unit 20 satisfies the requirement.
  • the data transmission unit 22 transmits the DNN generated by the DNN generation unit 20 and the verification data generated by the verification data generation unit 21 to the client 3.
  • data used for generating a DNN model and generating verification data is registered.
  • functional modules that realize various functions and learned DNNs are registered for each framework, and data sets corresponding to the functional modules are registered.
  • the DNN generation unit 20 reads data corresponding to the requirement from the DNN database 23, and generates a DNN that satisfies the requirement using the read data.
  • the verification data generation unit 21 generates verification data using the data read from the DNN database 23.
  • FIG. 1 shows a configuration in which the server 2 includes the DNN database 23, but the DNN database 23 may be provided in a device different from the server 2.
  • the DNN database 23 may be provided in a storage device on a cloud that can access data from the server 2.
  • the data verification unit 30 in the client 3 verifies whether the DNN satisfies the requirement, using the verification data received from the server 2.
  • the data verifying unit 30 verifies that the DNN satisfies the requirements, the DNN can be operated in various applications.
  • the request transmission unit 31 transmits a DNN generation request to the server 2.
  • the request transmission unit 31 transmits the request including the cause of the re-request and the verification result to the server 2 again.
  • the cause of the re-request is, for example, information indicating a function and performance that are not achieved by DNN among functions and performance set in advance.
  • the server 2 can finally generate a DNN that satisfies the requirements by repeating DNN learning in consideration of the cause of the re-request and the verification result.
  • FIG. 2 is a diagram showing an example of registered contents in the DNN database 23.
  • the DNN database 23 function modules for realizing various functions, data sets corresponding to the function modules, and learned DNNs are registered.
  • the data set includes teacher data used for DNN learning and test data used for DNN performance evaluation.
  • the learning module and the inference module must have the same framework. This means that when the frameworks are different, there are differences in the number of significant digits after the decimal point and the way of internal numerical calculation between the frameworks, and the presence or absence of function support.
  • the learning module 1A and the inference module 1A are implemented by a common framework (1), and the learning module 1A and the inference module 1A can use the learned DNN 1A.
  • the learning module 2A and the inference module 2A are implemented by a common framework (2), and the learned DNN2A-1 and the learned DNN2A-2 can be used.
  • an inference module implemented in a framework different from the learning module can use DNN.
  • the network description conversion module may be prepared independently, but may support an open neural network conversion format (ONNX) or a neural network conversion format (NNEF).
  • the DNN generation unit 20 can generate the learned DNN2A-1 and the learned DNN2A-2 that can be used by the inference module 2A by converting the network description format of the learned DNN1A by the network description conversion module. .
  • FIG. 3 is a flowchart showing the data processing method according to the first embodiment, and shows a series of processing by the server 2 and the client 3.
  • the DNN generation unit 20 generates a DNN learned so as to satisfy the requirements included in the command from the user in response to the request received from the client 3 (step ST1).
  • the requirements included in the instruction from the user include the function desired to be realized by DNN, the required performance, and the type of framework.
  • the hardware performance of the client 3 may be included in the command.
  • the hardware performance includes, for example, the memory capacity of the client 3 and the calculation accuracy of the processor.
  • the DNN generation unit 20 can generate a DNN model that can be operated with the hardware performance of the client 3.
  • the appropriate DNN processing characteristics differ depending on the inference function. For example, in the inference function of immediately recognizing the subject of an image after the image is taken, it is required that the delay of calculation processing in the client 3 is short. As described above, the DNN generation unit 20 may generate a DNN that matches the request of the inference operation characteristic of the client 3.
  • the DNN generation unit 20 may determine the structure of a convolutional neural network (CNN) using a learned filter. In addition, when there is a requirement to limit the DNN data size, the DNN generation unit 20 may perform a compression process on the DNN so as to satisfy the DNN data size request.
  • CNN convolutional neural network
  • the DNN generation unit 20 obtains a verification result that the DNN database 23 performs performance verification to satisfy the required performance when the learned DNN that is estimated to satisfy the requirements included in the instruction from the user exists in the DNN database 23. When this is done, this DNN is output to the verification data generator 21.
  • the DNN generation unit 20 learns the DNN using the DNN database 23. For example, the DNN generation unit 20 learns DNN using a module and a data set implemented by a framework that matches the instruction.
  • the DNN generation unit 20 may learn the DNN and convert it to a network description format that matches the command even if the framework that matches the command does not exist in the DNN database 23.
  • the DNN generation unit 20 verifies the performance of the DNN obtained by learning and obtains a verification result that satisfies the required performance, the DNN generation unit 20 outputs the DNN to the verification data generation unit 21.
  • the DNN generation unit 20 may perform selection of teacher data according to required performance.
  • the verification data generation unit 21 generates verification data used for verification of whether or not the DNN generated by the DNN generation unit 20 satisfies the requirement (step ST2).
  • the verification data is data corresponding to the DNN verification method performed by the client 3.
  • Examples of the DNN verification method include a method using an error back propagation method (hereinafter referred to as BP) and an intermediate description of DNN. There is a method using a child.
  • the verification data generation unit 21 updates the input value and the output value given to the DNN and the weight of the node of the DNN intermediate layer to which these are given by the BP.
  • the verification data including the gradient value and the threshold value related to the gradient value is generated.
  • the verification data generation unit 21 relates to the input value given to the DNN, the intermediate descriptor when the input value is given to the DNN, and the intermediate descriptor. Verification data including a threshold value is generated.
  • the data transmission unit 22 transmits the learned DNN generated by the DNN generation unit 20 and the verification data generated by the verification data generation unit 21 to the client 3 (step ST3).
  • the data transmission unit 22 generates transmission data in which header information such as a framework name is integrated with respect to the learned DNN and the verification data, and transmits the generated transmission data to the client 3.
  • the data verification unit 30 verifies whether or not the DNN generated by the DNN generation unit 20 satisfies the requirement using the verification data received from the server 2 (step ST4). For example, the data verification unit 30 uses the difference between the gradient value obtained by performing the weight update by BP on the learned DNN given the arbitrary input value and output value and the gradient value included in the verification data. Based on the result of comparison with the threshold included in the data, it is verified whether the DNN satisfies the requirement. In addition, the data verification unit 30 compares the difference between the intermediate descriptor obtained by giving an arbitrary input value to the learned DNN and the intermediate descriptor included in the verification data with the threshold included in the verification data. Based on the above, it is verified whether the DNN satisfies the requirement.
  • step ST4 When the verification result that the learned DNN satisfies the requirements is obtained (step ST4; YES), the data verification unit 30 ends the process of FIG. At this time, the learned DNN operates normally on the client 3 and can be operated by various applications.
  • the data verification unit 30 determines that DNN needs to be re-learned, and has learned the verified DNN.
  • the re-request factor and information indicating the details are output to the request transmission unit 31.
  • the request transmission unit 31 generates a request including information indicating the learned DNN, the cause of the re-request, and the details input from the data verification unit 30, and transmits the generated request to the server 2 (step ST5).
  • the server 2 executes the process from step ST1 again in response to the request received from the client 3.
  • a series of processes from step ST1 to step ST5 are repeatedly executed until a DNN that satisfies the requirements included in the user's command is obtained.
  • FIG. 4 is a flowchart showing the DNN generation process, and shows details of the process of step ST1 in FIG.
  • the DNN generation unit 20 receives an instruction including requirements required for the DNN (step ST1a).
  • the DNN generation unit 20 searches the DNN database 23 based on the type of framework included in the instruction, the target function, and the required performance, and the learned DNN estimated to satisfy the requirements included in the instruction is obtained. It is confirmed whether or not it exists (step ST2a).
  • the DNN generation unit 20 determines whether or not the learned DNN satisfies the required performance included in the instruction. This is verified (step ST3a). At this time, when a verification result is obtained that the learned DNN satisfies the required performance included in the instruction (step ST3a; YES), the DNN generation unit 20 determines the learned DNN, the DNN generation method, the achieved performance, and the required performance. The data is output to the verification data generation unit 21 (step ST4a).
  • the DNN generation method includes a method of using an existing DNN in the DNN database 23 or a method of learning a DNN.
  • a DNN learning method for example, there is a learning method using BP.
  • a data set used for DNN learning is also set.
  • the achieved performance is the DNN performance achieved by the DNN learning performed by the DNN generation unit 20.
  • the required performance is the DNN performance corresponding to the requirements included in the instruction received from the user.
  • step ST2a when the learned DNN estimated to satisfy the requirement included in the instruction does not exist in the DNN database 23 (step ST2a; NO), a verification result that the learned DNN does not satisfy the required performance included in the instruction is obtained. If it is determined (step ST3a; NO), the DNN generation unit 20 checks whether or not a framework that matches the framework included in the instruction exists in the DNN database 23 (step ST5a).
  • the DNN generation unit 20 When there is a framework in the DNN database 23 that matches the framework included in the instruction (step ST5a; YES), the DNN generation unit 20 reads data corresponding to this framework from the DNN database 23 and is included in the instruction. DNN is learned so as to satisfy the requirements (step ST6a). Subsequently, the DNN generation unit 20 verifies whether the learned DNN satisfies the required performance (step ST7a).
  • step ST7a when a verification result that the learned DNN satisfies the required performance included in the instruction is obtained (step ST7a; YES), the DNN generation unit 20 proceeds to step ST4a and generates the learned DNN and DNN and Requirements such as required performance are output to the verification data generation unit 21.
  • step ST7a NO
  • the DNN generation unit 20 returns to the process of step ST6a, and the DNN that satisfies the required performance is found. DNN learning is repeated until it is obtained.
  • the DNN generation unit 20 can convert to a network description format that matches the requirements included in the instruction. (Step ST8a). For example, the DNN generation unit 20 checks whether there is a conversion module in the DNN database 23 that performs conversion to a network description format that matches the requirements of the instruction.
  • step ST8a When conversion to a network description format matching the requirements included in the instruction is possible (step ST8a; YES), the DNN generation unit 20 learns the DNN so as to satisfy the requirements included in the instruction (step ST9a). Subsequently, after the DNN learning is performed, the DNN generation unit 20 performs conversion into a network description format that matches the requirements of the instruction (step ST10a). Thereafter, the DNN generation unit 20 verifies whether the learned DNN satisfies the required performance (step ST11a).
  • step ST11a When a verification result that the learned DNN satisfies the required performance included in the instruction is obtained (step ST11a; YES), the DNN generation unit 20 proceeds to step ST4a and generates the learned DNN, DNN generation method, required performance, and the like. Are output to the verification data generation unit 21.
  • step ST11a NO
  • step ST9a the DNN generation unit 20 returns to the process of step ST9a to obtain a DNN that satisfies the required performance. DNN learning is repeated until
  • a DNN generation unit 20 determines that it is impossible to generate a DNN that can be operated by the client 3 (step ST12a). At this time, the DNN generation unit 20 ends the process of FIG. 4 without generating a DNN.
  • FIG. 5 is a flowchart showing verification data generation processing by BP, and shows details of the processing in step ST2 in FIG.
  • the verification data generation unit 21 corresponds to the learned DNN, the DNN generation method, the DNN achievement performance achieved by the DNN learning performed by the DNN generation unit 20, and the requirements included in the instruction from the DNN generation unit 20.
  • the requested performance of DNN is input (step ST1b).
  • the verification data generation unit 21 selects an arbitrary input value and output value for the learned DNN (step ST2b).
  • the verification data generation unit 21 updates the weight of the node in the intermediate layer by BP on the learned DNN given the input value and output value selected in step ST2b (step ST3b). Subsequently, the verification data generation unit 21 calculates and stores the gradient value of the loss function when the weight of the intermediate layer node is updated (step ST4b). The verification data generation unit 21 determines a threshold value related to the gradient value calculated in step ST4b (step ST5b).
  • the verification data generation unit 21 uniquely determines a threshold value related to the gradient value based on the required performance and achieved performance of the DNN.
  • a threshold value related to the gradient value may be determined by repeating the DNN performance evaluation between the server 2 and the client 3 using the test data registered in the DNN database 23.
  • the verification data generation unit 21 outputs the learned DNN input from the DNN generation unit 20 and the verification data to the data transmission unit 22 (step ST6b).
  • the verification data includes the input value and output value selected in step ST2b, the gradient value calculated in step ST4b, and the threshold value determined in step ST5b.
  • the verification data generating unit 21 may generate verification data using the DNN intermediate descriptor.
  • FIG. 6 is a flowchart showing verification data generation processing using the DNN intermediate descriptor, and shows details of the processing in step ST2 in FIG.
  • the verification data generation unit 21 corresponds to the learned DNN, the DNN generation method, the DNN achievement performance achieved by the DNN learning performed by the DNN generation unit 20, and the requirements included in the instruction from the DNN generation unit 20.
  • the requested performance of DNN is input (step ST1c).
  • the verification data generation unit 21 selects an arbitrary input value for the learned DNN (step ST2c). For example, the verification data generation unit 21 selects teacher data or test data registered in the DNN database 23 or a value determined by a random number as an input value.
  • the verification data generation unit 21 calculates the output value of the DNN by giving the input value selected in Step ST2c to the learned DNN (Step ST3c).
  • the verification data generation unit 21 calculates and stores the intermediate descriptor of the node in the intermediate layer (step ST4c).
  • the verification data generation unit 21 determines a threshold value for the intermediate descriptor calculated in step ST4c (step ST5c).
  • the verification data generation unit 21 uniquely determines a threshold value related to the intermediate descriptor based on the required performance and achieved performance of the DNN.
  • a difference in performance between the server 2 and the client 3 can be allowed, so that a large value is set for the threshold for the intermediate descriptor.
  • the verification data generation unit 21 may determine a threshold value related to the intermediate descriptor by repeating the performance evaluation of the DNN between the server 2 and the client 3 using the test data registered in the DNN database 23.
  • the verification data generation unit 21 outputs the learned DNN input from the DNN generation unit 20 and the verification data to the data transmission unit 22 (step ST6c).
  • the verification data includes the input value selected in step ST2c, the intermediate descriptor calculated in step ST4c, and the threshold value determined in step ST5c.
  • FIG. 7 is a flowchart showing the data transmission process, and shows details of the process in step ST3 in FIG.
  • the data transmission unit 22 inputs transmission data from the verification data generation unit 21 (step ST1d).
  • the transmission data includes learned DNN input from the DNN generation unit 20 and verification data.
  • the data transmission unit 22 integrates the header information into the transmission data input in step ST1d (step ST2d).
  • the header information includes additional information such as the name of the framework corresponding to DNN, for example.
  • the data transmission unit 22 integrates the transmission data input in step ST1d (step ST3d). As a result, the learned DNN and verification data are integrated into one transmission data.
  • the data transmission unit 22 performs encryption processing on the transmission data integrated in step ST3d (step ST4d). For example, the data transmission unit 22 performs encryption that can be decrypted by the client 3 on the transmission data.
  • the data transmission unit 22 transmits the encrypted data to the client 3 (step ST5d).
  • the data transmission unit 22 may guarantee the consistency between the transmission data from the server 2 and the reception data of the client 3 using a hash code.
  • the data transmission unit 22 may add a function of detecting data falsification to the transmission data by generating transmission data for requesting a digital signature when received.
  • FIG. 8 is a flowchart showing the operation of the client 3 according to the first embodiment, and shows specific processing of step ST4 and step ST5 in FIG.
  • the data verification unit 30 receives the transmission data from the server 2 (step ST1e).
  • the transmission data from the server 2 includes the learned DNN generated by the DNN generation unit 20 and the verification data generated by the verification data generation unit 21.
  • the data verification unit 30 performs a decoding process on the received data (step ST2e), and confirms whether or not the inference process using the learned DNN extracted from the decoded data is possible in the client 3 (step ST3e). For example, when there is a difference between the framework corresponding to the learned DNN extracted from the decoded data and the framework used by the client 3 in the number of significant digits after the decimal point and the way of internal numerical calculation, the data verification unit 30 Determines that inference processing using the learned DNN is impossible in the client 3. The data verification unit 30 determines that inference processing using the learned DNN in the client 3 is impossible if the memory capacity of the client 3 does not satisfy the memory capacity necessary for the calculation of the DNN generated by the server 2. To do.
  • the data verification unit 30 uses the verification data received from the server 2 to check whether the learned DNN satisfies the performance requirement. It is verified whether or not (step ST4e).
  • the data verification unit 30 outputs the learned DNN to various applications (step ST5e). As a result, the learned DNN can be used by various applications.
  • step ST3e; NO when inference processing by the learned DNN is impossible in the client 3 (step ST3e; NO), or when the learned DNN does not satisfy the performance requirements (step ST4e; NO), the data verification unit 30 has learned The DNN and the cause of the re-request are output to the request transmission unit 31 (step ST6e).
  • the request transmission unit 31 transmits a request for generating a DNN that satisfies the same requirements as the learned DNN to the server 2 including the cause of the re-request.
  • FIG. 9 is a flowchart showing DNN performance verification processing by BP, and shows details of the processing in step ST4e in FIG.
  • the data verification unit 30 acquires learned DNN and verification data from the data received from the server 2 and decoded (step ST1f).
  • the verification data includes an input value and an output value for the learned DNN, a gradient value obtained by performing BP on the learned DNN that has been given these values, and a threshold value related to the gradient value. .
  • the data verification unit 30 updates the weight of the intermediate layer node by BP on the learned DNN given the arbitrary input value and output value (step ST2f), and updates the weight of the intermediate layer node. Calculate the slope value of the loss function. Subsequently, the data verification unit 30 calculates a difference between the gradient value calculated in step ST2f and the gradient value included in the verification data (step ST3f).
  • the data verification unit 30 checks whether or not the difference calculated in step ST3f is smaller than a threshold value included in the verification data (step ST4f). When the difference calculated in step ST3f is smaller than the threshold (step ST4f; YES), the data verification unit 30 determines that the learned DNN satisfies the performance requirement (step ST5f). On the other hand, when the difference calculated in step ST3f is equal to or greater than the threshold (step ST4f; NO), the data verification unit 30 outputs a verification result that the learned DNN does not satisfy the performance requirement to the request transmission unit 31 (step ST6f). ).
  • the difference between the gradient value obtained by performing weight update by BP on the learned DNN given arbitrary input values and output values and the gradient value included in the verification data is smaller than the threshold value.
  • the present invention is not limited to this. For example, depending on the threshold value determined by the verification data generation unit 21, when the difference is smaller than the threshold value, it is determined that the learned DNN does not satisfy the performance requirement, and when the difference is equal to or greater than the threshold value, the learned DNN is It may be determined that the performance requirement is satisfied.
  • FIG. 10 is a flowchart showing the DNN performance verification process using the DNN intermediate descriptor, and shows details of the process in step ST4e in FIG.
  • the data verification unit 30 acquires learned DNN and verification data from the data received from the server 2 and decoded (step ST1g). It is assumed that the verification data includes an input value for the learned DNN, an intermediate descriptor obtained from the learned DNN that has given this value, and a threshold value related to the intermediate descriptor.
  • the data verification unit 30 gives an arbitrary input value to the learned DNN (step ST2g), and calculates a difference between the intermediate descriptor obtained from the intermediate layer and the intermediate descriptor included in the verification data (step ST3g). ). The data verification unit 30 checks whether or not the difference calculated in step ST3g is smaller than a threshold value included in the verification data (step ST4g).
  • step ST3g When the difference calculated in step ST3g is smaller than the threshold (step ST4g; YES), the data verification unit 30 determines that the learned DNN satisfies the performance requirement (step ST5g). On the other hand, when the difference calculated in step ST3g is equal to or greater than the threshold (step ST4g; NO), the data verification unit 30 outputs a verification result that the learned DNN does not satisfy the performance requirement to the request transmission unit 31 (step ST6g). ).
  • the verification result that the learned DNN does not satisfy the performance requirement includes the learned DNN, the cause of the re-request, and the detailed contents thereof.
  • the learned DNN when the difference between the intermediate descriptor obtained by giving an arbitrary input value to the learned DNN and the intermediate descriptor included in the verification data is smaller than the threshold, the learned DNN is a performance requirement. Although it determined with satisfy
  • FIG. 11 is a flowchart showing the request retransmission process, and shows details of the process in step ST6e in FIG.
  • the request transmission unit 31 inputs request data from the data verification unit 30 (step ST1h).
  • the request data includes the learned DNN determined not to satisfy the performance requirement, the cause of the re-request, and the detailed contents thereof.
  • the request transmission unit 31 generates request information using the request data input from the data verification unit 30 (step ST2h). For example, the request transmission unit 31 generates request information that integrates the learned DNN, the cause of the re-request, and the detailed contents thereof.
  • the request transmission unit 31 performs an encryption process on the request information generated in step ST2h (step ST3h). For example, the request transmission unit 31 performs encryption that can be decrypted by the server 2 on the request information. Next, the request transmission unit 31 transmits request information including the cause and details of the re-request to the server 2 (step ST4h).
  • the server 2 includes a processing circuit for executing the processing from step ST1 to step ST3 in FIG.
  • This processing circuit may be dedicated hardware, or may be a CPU (Central Processing Unit) that executes a program stored in the memory.
  • the functions of the data verification unit 30 and the request transmission unit 31 in the client 3 are realized by a processing circuit.
  • the client 3 includes a processing circuit for executing the processing from step ST4 to step ST5 in FIG.
  • the processing circuit may be dedicated hardware, or may be a CPU that executes a program stored in a memory.
  • FIG. 12A is a block diagram showing a hardware configuration for realizing the functions of the server 2.
  • FIG. 12B is a block diagram illustrating a hardware configuration for executing software that implements the functions of the server 2.
  • the communication device 100 is a device that communicates with the client 3.
  • the DNN generation unit 20 receives a request from the request transmission unit 31 of the client 3 using the communication device 100, and the data transmission unit 22 transmits data to the data verification unit 30 of the client 3 using the communication device 100.
  • the DB interface 101 is an interface that relays data exchange between the DNN database 102, the DNN generation unit 20, and the verification data generation unit 21.
  • the DNN database 102 is registered in the DNN database 102, which is the DNN database 23 shown in FIG. 1, and is used to generate a DNN model and verification data.
  • the DNN database 102 may be provided independently of the server 2. For example, it may be provided in a storage device that exists on a cloud that can access data from the server 2.
  • FIG. 12C is a block diagram showing a hardware configuration for realizing the function of the client 3.
  • FIG. 12D is a block diagram illustrating a hardware configuration for executing software that implements the functions of the client 3.
  • the communication device 106 is a device that communicates with the server 2.
  • the data verification unit 30 receives data from the data transmission unit 22 of the server 2 using the communication device 106, and the request transmission unit 31 transmits a request to the DNN generation unit 20 of the server 2 using the communication device 106.
  • the processing circuit 103 is, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), or the like. ), FPGA (Field-Programmable Gate Array), or a combination thereof.
  • the functions of the DNN generation unit 20, the verification data generation unit 21, and the data transmission unit 22 in the server 2 may be realized by separate processing circuits, or these functions may be realized by a single processing circuit.
  • the processing circuit is the processor 104 shown in FIG. 12B
  • the functions of the DNN generation unit 20, the verification data generation unit 21, and the data transmission unit 22 in the server 2 are realized by software, firmware, or a combination of software and firmware.
  • the software or firmware is described as a program and stored in the memory 105.
  • the processor 104 reads out and executes the program stored in the memory 105, thereby realizing the functions of the DNN generation unit 20, the verification data generation unit 21, and the data transmission unit 22 in the server 2. That is, the server 2 includes a memory 105 for storing a program that, when executed by the processor 104, results in the processing from step ST1 to step ST3 shown in FIG. These programs cause the computer to execute the procedures or methods of the DNN generation unit 20, the verification data generation unit 21, and the data transmission unit 22.
  • the memory 105 may be a computer-readable storage medium storing a program for causing a computer to function as the DNN generation unit 20, the verification data generation unit 21, and the data transmission unit 22.
  • the processing circuit 107 is, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or the like. This is a combination.
  • the functions of the data verification unit 30 and the request transmission unit 31 in the client 3 may be realized by separate processing circuits, or these functions may be realized by a single processing circuit.
  • the processing circuit is the processor 108 shown in FIG. 12D
  • the functions of the data verification unit 30 and the request transmission unit 31 in the client 3 are realized by software, firmware, or a combination of software and firmware.
  • the software or firmware is described as a program and stored in the memory 109.
  • the processor 108 implements the functions of the data verification unit 30 and the request transmission unit 31 in the client 3 by reading and executing the program stored in the memory 109. That is, the client 3 includes a memory 109 for storing a program in which the processing from step ST4 to step ST5 shown in FIG. These programs cause the computer to execute the procedures or methods of the data verification unit 30 and the request transmission unit 31.
  • the memory 109 may be a computer-readable storage medium that stores a program for causing a computer to function as the data verification unit 30 and the request transmission unit 31.
  • the memory 105 or 109 includes, for example, RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically-EPROM) or non-volatile.
  • RAM Random Access Memory
  • ROM Read Only Memory
  • flash memory EPROM (Erasable Programmable Read Only Memory)
  • EEPROM Electrically-EPROM
  • a part of the functions of the DNN generation unit 20, the verification data generation unit 21, and the data transmission unit 22 may be realized by dedicated hardware, and a part may be realized by software or firmware.
  • the DNN generation unit 20 realizes the function by the processing circuit 103 that is dedicated hardware, and the verification data generation unit 21 and the data transmission unit 22 read and execute the program stored in the memory 105 by the processor 104. To realize the function.
  • the processing circuit can realize the above functions by hardware, software, firmware, or a combination thereof.
  • part of the functions of the data verification unit 30 and the request transmission unit 31 may be realized by dedicated hardware, and a part may be realized by software or firmware.
  • the data verification unit 30 realizes the function by the processing circuit 107 that is dedicated hardware
  • the request transmission unit 31 realizes the function by the processor 108 reading and executing the program stored in the memory 109.
  • the processing circuit can realize the above functions by hardware, software, firmware, or a combination thereof.
  • the server 2 generates verification data used for verifying whether or not the DNN satisfies the requirement and transmits the data to the client 3, thereby simplifying the DNN verification process. be able to.
  • the software used between the server 2 and the client 3 is inconsistent, the DNN learned by the server 2 and the DNN requested by the client 3 cannot achieve the same performance.
  • the operating system OS
  • the software version is different, the same inference performance cannot be obtained between the DNN learned by the server 2 and the DNN requested by the client 3. Can be a factor.
  • the DNN learned by the server 2 and the DNN requested by the client 3 cannot achieve the same performance. For example, if the memory capacity of the client 3 that performs the inference process is less than the memory capacity required for the operation of the DNN generated by the server 2, the inference process cannot be performed on the client 3.
  • the client 3 verifies whether or not the requested requirement is satisfied by the DNN, designs the DNN again based on the verification result, and sends the DNN to the server 2 so as to satisfy the requirement according to this design. Need to be re-learned.
  • the server 2 generates verification data and transmits it to the client 3. Therefore, the client 3 does not need to prepare all the verification data in the DNN performance verification. The verification process can be simplified.
  • the client 3 uses the verification data received from the server 2 to verify whether or not the learned DNN generated by the server 2 satisfies the requirements, so that the DNN verification process is simplified. It can be made.
  • the data processing system 1 according to the first embodiment includes the server 2 and the client 3, the same effects as described above can be obtained. Furthermore, in the data processing method according to the first embodiment, since the server 2 and the client 3 operate as shown in FIG. 3, the same effects as described above can be obtained.
  • the learning apparatus can simplify the DNN verification process, it can be used in, for example, a computer system that recognizes a photographed image around the vehicle.
  • 1 data processing system 1 data processing system, 2 servers, 3 clients, 20 DNN generation unit, 21 verification data generation unit, 22 data transmission unit, 23,102 DNN database, 30 data verification unit, 31 request transmission unit, 100, 106 communication device, 101 DB interface, 103, 107 processing circuit, 104, 108 processor, 105, 109 memory.

Abstract

Un serveur (2) génère des données de vérification servant à vérifier si un réseau neuronal profond répond à une exigence et les transmet à un client (3).
PCT/JP2018/017538 2018-05-02 2018-05-02 Dispositif d'apprentissage, dispositif de vérification, système et procédé de traitement de données WO2019211909A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/JP2018/017538 WO2019211909A1 (fr) 2018-05-02 2018-05-02 Dispositif d'apprentissage, dispositif de vérification, système et procédé de traitement de données
JP2020517000A JP6742565B2 (ja) 2018-05-02 2018-05-02 学習装置、検証装置、データ処理システムおよびデータ処理方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2018/017538 WO2019211909A1 (fr) 2018-05-02 2018-05-02 Dispositif d'apprentissage, dispositif de vérification, système et procédé de traitement de données

Publications (1)

Publication Number Publication Date
WO2019211909A1 true WO2019211909A1 (fr) 2019-11-07

Family

ID=68386382

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2018/017538 WO2019211909A1 (fr) 2018-05-02 2018-05-02 Dispositif d'apprentissage, dispositif de vérification, système et procédé de traitement de données

Country Status (2)

Country Link
JP (1) JP6742565B2 (fr)
WO (1) WO2019211909A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20220069738A (ko) * 2020-11-20 2022-05-27 (주)한국플랫폼서비스기술 쿼리 기반 딥러닝 추론 시스템을 이용한 공작 기계 예지 보전 시스템 및 그 방법
WO2022190801A1 (fr) * 2021-03-10 2022-09-15 ソニーセミコンダクタソリューションズ株式会社 Dispositif de traitement d'informations, système de traitement d'informations, procédé de traitement d'informations, et support d'enregistrement

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002342739A (ja) * 2001-05-17 2002-11-29 Kddi Corp 通信ネットワークを介したニューラルネットワーク処理システム及びそのプログラムを格納したプログラム記憶媒体
JP2017059031A (ja) * 2015-09-17 2017-03-23 日本電気株式会社 情報処理装置、情報処理方法、及び、プログラム
JP2017174298A (ja) * 2016-03-25 2017-09-28 株式会社デンソーアイティーラボラトリ ニューラルネットワークシステム、端末装置、管理装置およびニューラルネットワークにおける重みパラメータの学習方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002342739A (ja) * 2001-05-17 2002-11-29 Kddi Corp 通信ネットワークを介したニューラルネットワーク処理システム及びそのプログラムを格納したプログラム記憶媒体
JP2017059031A (ja) * 2015-09-17 2017-03-23 日本電気株式会社 情報処理装置、情報処理方法、及び、プログラム
JP2017174298A (ja) * 2016-03-25 2017-09-28 株式会社デンソーアイティーラボラトリ ニューラルネットワークシステム、端末装置、管理装置およびニューラルネットワークにおける重みパラメータの学習方法

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20220069738A (ko) * 2020-11-20 2022-05-27 (주)한국플랫폼서비스기술 쿼리 기반 딥러닝 추론 시스템을 이용한 공작 기계 예지 보전 시스템 및 그 방법
KR102568010B1 (ko) * 2020-11-20 2023-08-22 (주)한국플랫폼서비스기술 쿼리 기반 딥러닝 추론 시스템을 이용한 공작 기계 예지 보전 시스템 및 그 방법
WO2022190801A1 (fr) * 2021-03-10 2022-09-15 ソニーセミコンダクタソリューションズ株式会社 Dispositif de traitement d'informations, système de traitement d'informations, procédé de traitement d'informations, et support d'enregistrement

Also Published As

Publication number Publication date
JPWO2019211909A1 (ja) 2020-08-20
JP6742565B2 (ja) 2020-08-19

Similar Documents

Publication Publication Date Title
US20230401445A1 (en) Multi-domain joint semantic frame parsing
US10268679B2 (en) Joint language understanding and dialogue management using binary classification based on forward and backward recurrent neural network
US10747505B1 (en) API specification generation
WO2018044633A1 (fr) Apprentissage de bout en bout d'agents de dialogue pour accès à des informations
KR20190038923A (ko) 사용자 신원 검증 방법, 장치 및 시스템
US11562245B2 (en) Neural network model generation and distribution with client feedback
WO2023030348A1 (fr) Procédé et appareil de génération d'image, dispositif et support de stockage
CN106453474A (zh) 在不稳定网络环境中的大文件的网络传输
WO2019211909A1 (fr) Dispositif d'apprentissage, dispositif de vérification, système et procédé de traitement de données
CN111046027A (zh) 时间序列数据的缺失值填充方法和装置
WO2019185981A1 (fr) Génération ou obtention d'un réseau neuronal mis à jour
CN113473149A (zh) 用于无线图像传输的语义信道联合编码方法及装置
JP2018197832A (ja) ブロックチェーン更新システム、サーバ装置、クライアント装置、ブロックチェーン更新方法、およびプログラム
WO2023051238A1 (fr) Procédé et appareil pour générer une figure d'animal, dispositif, et support de stockage
WO2022246986A1 (fr) Procédé, appareil et dispositif de traitement de données, et support de stockage lisible par ordinateur
WO2023151333A1 (fr) Procédé et appareil de traitement vidéo, dispositif, et support de stockage
CN116743785A (zh) 基于雾计算的云网数据存储方法、装置、设备及介质
CN116489621A (zh) 车钥匙的分享方法、装置、设备及介质
WO2023014298A2 (fr) Procédé et appareil de construction de réseau neuronal
CN113112269B (zh) 多重签名方法、计算机设备和存储介质
US20220327526A1 (en) Method for enabling efficient evaluation of transactions in a distributed ledger network
CN113628052A (zh) 基于预言机的区块链资产与合约处理方法、系统及装置
US20170248916A1 (en) Method and system for image processing and data transmission in network-based multi-camera environment
US20230176902A1 (en) System and method for automated onboarding
US20240135207A1 (en) Method and system for computer vision inferencing using a processing system with dedicated hardware

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: 18917380

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020517000

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18917380

Country of ref document: EP

Kind code of ref document: A1