Systems, Methods, and Media for Managing Prediction Models

Abstract

The present invention provides a system, a method, and a medium for managing a prediction model. The system manages a prediction model generated by using input data including a plurality of parameters in a learning process and includes: a first calculation unit configured to calculate a learning contribution rate of each of the plurality of parameters according to a learning result; a first specifying unit configured to specify a parameter with a low learning contribution rate according to the learning contribution rate calculated by the first calculation unit; a first instruction unit configured to give a relearning instruction to use input data excluding the parameter specified by the first specifying unit; a second specifying unit configured to specify a structure corresponding to the parameter specified by the first specifying unit; and a first issuing unit configured to issue a stop command for the structure specified by the second specifying unit.

Description

Technical Field

[0001] The present invention relates to a system, a method, and a medium for managing a prediction model. Background Art

[0002] In machine learning, a prediction model is learned by understanding features from input data for learning, and the fully learned prediction model is used for actual operation. In the prior art, a system is known that predicts a specific event by using prediction input data using such a prediction model.

[0003] In a system that performs such prediction, in order to improve prediction accuracy by making predictions using various types of input data, the input data is usually a multi-dimensional vector. Incidentally, when the dimension of the input data is large, in addition to the huge computational learning cost, there are the following problems: a large amount of learning data is required to appropriately learn features. It is well known that when the dimension of a multi-dimensional vector is large, the learning accuracy actually deteriorates. This is because, due to the increase in dimension, the width of the features taken by the multi-dimensional vector expands exponentially, and thus the features cannot be sufficiently understood from a limited amount of input data.

[0004] Therefore, in the prior art, research has been conducted on reducing the dimension of learning input data. For example, Japanese Patent Application Laid-Open No. 2011-197934 discloses an invention in which the dimension reduction process includes setting parameters with a low contribution degree to learning (hereinafter referred to as learning contribution rate) as objects for dimension reduction.

[0005] In order to appropriately understand and learn features from multi-dimensional learning input data, as described above, it is necessary to combine a dimension reduction process. However, in Japanese Patent Application Laid-Open No. 2011-197934, the dimension reduction process only does not use the parameters for dimension reduction. Therefore, when acquiring, transmitting, etc. the parameters for dimension reduction collected in each configuration (for example, a device, a sensor, or software) before reduction, the parameters for dimension reduction will continue to be collected even after the dimension reduction process. In this case, there is a problem that the operation costs of the configuration for acquiring the parameters for dimension reduction and the communication process that occurs are wasted. Summary of the Invention

[0006] An object of the present invention is to reduce the operation costs wasted in a system that makes predictions in machine learning.

[0007] According to an embodiment of the present invention, a system manages a prediction model generated by using input data including a plurality of parameters in learning processing and includes: a first calculation unit configured to calculate a learning contribution rate of each of the plurality of parameters according to a learning result; a first specifying unit configured to specify a parameter with a low learning contribution rate according to the learning contribution rate calculated by the first calculation unit; a first instruction unit configured to give a relearning instruction to use input data excluding the parameter specified by the first specifying unit; a second specifying unit configured to specify a configuration corresponding to the parameter specified by the first specifying unit; and a first issuing unit configured to issue a stop command for the configuration specified by the second specifying unit.

[0008] Other features of the present invention will become clear from the following description of exemplary embodiments with reference to the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 is a block diagram showing the configuration of an information processing system according to Example 1 of the present invention.

[0010] Figure 2A is a block diagram showing the hardware configuration of a parameter optimization system 101 and a gateway device 102.

[0011] Figure 2B is a block diagram showing the hardware configuration of a parameter acquisition device 103.

[0012] Figure 3 is a block diagram showing the functional configuration of a parameter optimization system 101.

[0013] Figure 4 is a block diagram showing the functional configuration of a gateway device 102.

[0014] Figure 5 is a block diagram showing the functional configuration of a parameter acquisition device 103.

[0015] Figure 6A is a sequence diagram according to Example 1 of the present invention.

[0016] Figure 6B is a flowchart showing a processing flow in a learning contribution determination unit 304 according to Example 1 of the present invention.

[0017] Figure 7A is a sequence diagram according to Example 2 of the present invention.

[0018] Figure 7B is a flowchart showing a processing flow in a learning contribution determination unit 304 according to Example 2 of the present invention.

[0019] Figure 8A It is a sequence diagram according to Example 3 of the present invention.

[0020] Figure 8B It is a flowchart showing the processing flow in the learning contribution determination unit 304 according to Example 3 of the present invention. Detailed implementation manners

[0021] Hereinafter, modes for implementing the present invention will be described in detail with reference to the drawings.

[0022] [Example 1]

[0023] Figure 1 It is a block diagram showing the configuration of the information processing system according to Example 1 of the present invention.

[0024] The information processing system 100 according to this example is a system including one or more parameter acquisition devices 103 and a machine learning system that learns and makes predictions using the parameters acquired by the parameter acquisition devices 103. In this example, it is assumed that the machine learning system is included in the parameter optimization system 101. In this example, a parameter acquisition device 103 is set for each parameter to be acquired. The information processing system 100 is a system that manages a prediction model.

[0025] As a specific example of the learning and prediction processing explored in this example, the autonomous driving of a car can be exemplified.

[0026] The information processing system 100 according to this example assigns the driver's behavior as an answer label to the parameters collected in various configurations installed in the car during manual driving, and generates supervised data, that is, learning input data. The various configurations installed in the car are parameter acquisition devices 103 such as image sensors, distance sensors, and acceleration sensors. Table 1 is a table showing an example of the supervised data. The information processing system 100 generates a prediction model by preparing and learning a lot of supervised data as shown in Table 1.

[0027] [Table 1]

[0028]

[0029] In actual operation, the information processing system 100 determines a route based on the current value coordinates acquired by the GPS, the destination coordinates input by the user, and the road information. GPS is the abbreviation of the Global Positioning System. The information processing system 100 travels along the determined route by inputting the real-time parameters collected by the parameter acquisition device 103 into the prediction model, predicting (determining) the next behavior, and operating the car. Table 2 is a table showing examples of the parameters and prediction results during prediction.

[0030] [Table 2]

[0031]

[0032] For example, applying the present invention to the autonomous driving of an automobile in this example is merely an example of the present invention. The present invention can be applied to any one of, for example, a smart home, an airplane, a ship, a robot such as a drone, an image processing device including at least one of a printing device and a scanner, a 3D printer, and a communication relay device including a plurality of parameter acquisition devices. That is, the prediction model according to the present invention can be applied to any control of a smart home, an airplane, a ship, a robot such as a drone, an image processing device including at least one of a printing device and a scanner, a 3D printer, and a communication relay device.

[0033] The information processing system 100 is configured by connecting the parameter optimization system 101 to the gateway device 102 via the network 104. The network 104 is, for example, an HTTP-based network. HTTP is an abbreviation for the HyperText Transfer Protocol. The parameter acquisition device 103 uses a measuring device to digitize and acquire various physical quantities. The gateway device 102 serves as a firewall. In addition, the gateway device 102 performs operation control on the parameter acquisition device 103. For example, the gateway device 102 selects the transmission via the network 104 of the parameters collected by the parameter acquisition device 103.

[0034] In this example, examples of the parameter acquisition device 103 include an image sensor, a distance sensor, an acceleration sensor, and a GPS terminal installed in an automobile. The gateway device 102 is installed in the automobile and performs operation control on the parameter acquisition device 103, transmission control of the parameters collected by the parameter acquisition device 103, and the like. An example of the gateway device 102 is an IoT control module connected to the network 104. IoT is an abbreviation for the Internet of Things. The gateway device 102 periodically or aperiodically sends the parameter values acquired by the parameter acquisition device 103 to the parameter optimization system 101 via the network 104. The parameter optimization system 101 performs learning processing and prediction processing based on the received parameter values. The parameter optimization system 101 can send a control command for the parameter acquisition device 103 to the gateway device 102 via the network 104.

[0035] Figure 2A is a block diagram showing the hardware configurations of the parameter optimization system 101 and the gateway device 102. In this example, the hardware configurations of the parameter optimization system 101 and the gateway device 102 are the same, so reference will be made to Figure 2ADescribe two hardware configurations. RAM 201 is a memory for temporarily recording data. ROM 202 is a memory for recording embedded programs and data. Network interface 203 is an interface for communicating with other computers and network devices connected to network 104 via network 104. The communication scheme for network interface 203 and network 104 can be either a wired scheme or a wireless scheme. The secondary storage device 205 is a storage device typified by an HDD or a flash memory. CPU 200 executes programs read from RAM 201, ROM 202, secondary storage device 205, etc. CPU 200, RAM 201, ROM 202, network interface 203, and secondary storage device 205 are interconnected via system bus 204. Unless otherwise stated, system bus 204 in this example is a bus that propagates control commands from CPU 200 to the hardware connected to system bus 204.

[0036] Figure 2B is a block diagram showing the hardware configuration of parameter acquisition device 103. Network interface 206 is an interface for communicating with other computers and network devices connected to network 104 via network 104. The communication scheme for network interface 203 and network 104 can be either a wired scheme or a wireless scheme. Parameter acquisition device core 207 measures and digitizes various physical quantities in the real world. The parameters measured and digitized by parameter acquisition device core 207 are sent via network 104 to other computers and network devices connected to network 104 through network interface 206.

[0037] Figure 3 is a block diagram showing the functional configuration of parameter optimization system 101. The functional configuration of parameter optimization system 101 is realized by having CPU 200 execute programs stored in RAM201, ROM 202, secondary storage device 205, etc.

[0038] First, the processing flow during learning will be described. The collected data receiving unit 300 of parameter optimization system 101 receives the parameters collected by parameter acquisition device 103 and sent via gateway device 102 as collected data. The collected data includes multiple parameters. The input data generation unit 301 preprocesses the collected data received by the collected data receiving unit 300 and processes the collected data to obtain input data. Examples of the collected data before preprocessing and the input data after preprocessing are shown in Table 3.

[0039] [Table 3]

[0040]

[0041] The parameter optimization system 101 stores the input data processed through preprocessing in the input data storage unit 302. The learning processing unit 303 reads the input data from the input data storage unit 302 and performs learning processing. The learning processing unit 303 generates a prediction model through the learning processing. In this example, it is assumed that a multi-layer neural network is used in the machine learning algorithm. Here, this is exemplary, and the present invention can also be applied even when other machine learning algorithms are used.

[0042] The learning contribution determination unit 304 calculates the learning contribution rate of each parameter included in the collected data according to the prediction model generated through the learning processing in the learning processing unit 303. In the calculation of the learning contribution rate, various schemes can be used by utilizing the machine learning algorithm. However, when taking a 3-layer neural network that processes N-dimensional input data at this time as an example for description, the learning contribution rate of the nth-dimensional parameter in the input data can be obtained in the following calculation expression (1). W ij Indicates the weight from node i of the pth layer to node j of the p+1th layer. X, Y, and M are the number of nodes in the intermediate layer (the first layer), the intermediate layer (the second layer), and the output layer (the third layer).

[0043]

[0044] The learning contribution determination unit 304 determines the parameters with a low learning contribution rate (hereinafter referred to as low contribution parameters) according to the calculated learning contribution rate. That is, the learning contribution determination unit 304 designates the parameters with a low learning contribution rate according to the calculated learning contribution rate. An example of the low contribution parameters and the learning contribution rate calculated by the learning contribution determination unit 304 according to the learning result is shown in Table 4. In Table 4, each of the first image sensor, the second image sensor, the first distance sensor, the second distance sensor, the third distance sensor, and the acceleration sensor is an example of the parameter acquisition device 103. Each of the first image sensor and the second image sensor is a device that acquires multiple parameters. Each of the first distance sensor, the second distance sensor, the third distance sensor, and the acceleration sensor is a device that acquires one parameter. In this example, for example, a learning contribution rate of 0.0005 is set as the threshold, the parameters with a learning contribution rate less than the threshold are set as low contribution parameters, and the parameters with a learning contribution rate equal to or greater than the threshold are not set as low contribution parameters. In this case, in Table 4, the parameters acquired by the second distance sensor (learning contribution rate of 0.0002) and the third distance sensor (learning contribution rate of 0.0004) are low contribution parameters.

[0045] [Table 4]

[0046]

[0047] The parameter optimization system 101 sends information about the low contribution parameters determined by the learning contribution determination unit 304 to the learning processing unit 303. The control command sending unit 305 sends an operation stop command for the parameter acquisition device 103 to the gateway device 102, and the parameter acquisition device 103 acquires the parameters considered to be low contribution parameters by the learning contribution determination unit 304. In the case of Table 4, the control command sending unit 305 sends an operation stop command for the second distance sensor and the third distance sensor to the gateway device 102.

[0048] Incidentally, in this example, for the first image sensor and the second image sensor, for example, a plurality of pixel data are acquired by one sensor, and the acquired plurality of pixel data are set as a plurality of parameters. In this way, when a plurality of parameters are acquired from one parameter acquisition device 103, the plurality of acquired parameters are considered to include parameters with a learning contribution rate less than the threshold and parameters with a learning contribution rate equal to or greater than the threshold.

[0049] Therefore, in this example, when a plurality of parameters are acquired from one parameter acquisition device 103 and the learning contribution rates of all the plurality of parameters are less than the threshold, the learning contribution determination unit 304 sets the plurality of parameters as low contribution parameters. When a plurality of parameters are acquired from one parameter acquisition device 103 and the learning contribution rate of any one of the plurality of parameters is equal to or greater than the threshold, the learning contribution determination unit 304 does not set the plurality of parameters as low contribution parameters.

[0050] In this example, when a plurality of parameters are acquired from one parameter acquisition device 103 and the sum value of the learning contribution rates of the plurality of parameters is less than the sum value threshold, the learning contribution determination unit 304 sets the plurality of parameters as low contribution parameters. When a plurality of parameters are acquired from one parameter acquisition device 103 and the sum value of the learning contribution rates of the plurality of parameters is equal to or greater than the sum value threshold, the learning contribution determination unit 304 does not set the plurality of parameters as low contribution parameters.

[0051] As described above, the control command sending unit 305 sends an operation stop command for the parameter acquisition device 103, and the parameter acquisition device 103 acquires a plurality of parameters considered to be low contribution parameters by the learning contribution determination unit 304.

[0052] When a plurality of parameters are acquired from one parameter acquisition device 103 and the learning contribution rate of any one of the plurality of parameters is less than the threshold, the learning contribution determination unit 304 may set the plurality of parameters as low contribution parameters. In this case, when a plurality of parameters are acquired from one parameter acquisition device 103 and the learning contribution rates of all the plurality of parameters are equal to or greater than the threshold, the learning contribution determination unit 304 may set the plurality of parameters as low contribution parameters.

[0053] After receiving information on the low contribution parameter from the learning contribution determination unit 304, the learning processing unit 303 acquires input data from the input data storage unit 302, excludes the low contribution parameter from the acquired input data, and performs the learning processing again. The parameter optimization system 101 stores the prediction model generated by the learning processing in the prediction model storage unit 306.

[0054] Next, the processing flow during prediction will be described. First, the parameter acquisition device 103 collects parameters. Subsequently, the parameter acquisition device 103 sends the parameters to the parameter optimization system 101 via the gateway device 102. The parameter optimization system 101 causes the collected data receiving unit 300 to receive the parameters from the gateway device 102 as collected data. The input data generation unit 301 preprocesses the collected data received by the collected data receiving unit 300 to process it into input data. The parameter optimization system 101 stores the input data processed by the input data generation unit 301 in the input data storage unit 302. The prediction processing unit 307 acquires the prediction model from the prediction model storage unit 306 and inputs the input data acquired from the input data storage unit 302 into the prediction model. The prediction processing unit 307 sends the prediction result output by the prediction model to the output unit 308. The output unit 308 outputs the prediction result from the prediction processing unit 307. The output processing performed by the output unit 308 can be any one of the following processes: displaying the prediction result on the screen, outputting the prediction result as sound, sending the prediction result via email or chat, storing the prediction result in a storage device, causing a printing device to print the prediction result, and performing control to send the prediction result to a device. The output processing performed by the output unit 308 can be a combination of the above processes. Examples of the prediction result and the parameters collected by the parameter acquisition device 103 are shown in Table 2.

[0055] Figure 4 is a block diagram showing the functional configuration of the gateway device 102. The functional configuration of the gateway device 102 is realized by causing the CPU 200 to execute programs stored in the RAM 201, ROM 202, secondary storage device 205, etc.

[0056] In the gateway device 102, the parameter receiving unit 404 periodically or at a predetermined timing receives parameters sent from the parameter acquisition device 103. The gateway device 102 sends the received parameters from the parameter sending unit 402 to the parameter optimization system 101 via the parameter sending control unit 403. In the gateway device 102, the control command receiving unit 400 receives a control command sent from the parameter optimization system 101, and the parameter acquisition device control unit 401 controls the parameter acquisition device 103 according to the received control command. For example, when the control command receiving unit 400 receives an operation stop command for a specific parameter acquisition device 103 among multiple parameter acquisition devices 103, the parameter acquisition device control unit 401 performs control to stop the operation of the specific parameter acquisition device 103.

[0057] Figure 5 is a block diagram showing the functional configuration of the parameter acquisition device 103. The functional configuration of the parameter acquisition device 103 is implemented by causing the CPU of the parameter acquisition device 103 to execute a program and / or by hardware. The operation of the parameter acquisition device 103 is implemented by sending the parameters collected by the parameter acquisition device core 207 from the network interface 206. The parameter acquisition device 103 sends the parameters collected by the parameter acquisition unit 501 from the parameter sending unit 500 to the gateway device 102.

[0058] Figure 6A is a sequence diagram according to Example 1 of the present invention. Figure 6B is a flowchart showing the processing flow in the learning contribution determination unit 304 according to Example 1 of the present invention. In Figure 6A in order to enhance visibility, the control command sending unit 305 and the control command receiving unit 400 that do not perform processing other than sending and receiving processing are omitted.

[0059] The learning processing unit 303 obtains input data for calculating the learning contribution rate from the input data storage unit 302 ( Figure 6A step S600). The learning processing unit 303 performs learning processing using the input data obtained in step S600 ( Figure 6A step S601), and sends the learning result to the learning contribution determination unit 304 ( Figure 6A step S602).

[0060] The learning contribution determination unit 304 receives the learning result sent in step S602, obtains the learning contribution rate of the parameters including the input data from the received learning result, and sets the parameters with a low learning contribution rate as low contribution parameters ( Figure 6A and Figure 6BStep S603). That is, the learning contribution determination unit 304 designates parameters with a low learning contribution rate. Based on the determination information of the low contribution parameters in step S603, the learning contribution determination unit 304 sends a relearning command excluding the low contribution parameters to the learning processing unit 303( Figure 6A and Figure 6B Step S604). That is, the learning contribution determination unit 304 gives an instruction to perform relearning using the input data excluding the parameters designated as low contribution parameters.

[0061] The learning processing unit 303 receives the relearning command sent in step S604, and in response to the received relearning command, obtains learning input data excluding the low contribution parameters from the input data storage unit 302( Figure 6A Step S605). The learning processing unit 303 uses the learning input data obtained in step S605 to perform learning processing( Figure 6A Step S606). The learning processing unit 303 stores the prediction model generated by the learning processing in step S606 in the prediction model storage unit 306, and replaces the prediction model during actual operation with this prediction model through automatic or manual operation.

[0062] The learning contribution determination unit 304 determines the parameter acquisition device 103 that acquires parameters considered to be low contribution parameters( Figure 6A and Figure 6B Step S607). That is, the learning contribution determination unit 304 designates in step S607 the parameter acquisition device 103 that acquires parameters considered to be low contribution parameters. The parameter acquisition device 103 that acquires parameters considered to be low contribution parameters is an example of a configuration corresponding to the designated parameters. The learning contribution determination unit 304 sends a stop command for the operation of the parameter acquisition device 103 designated in step S607 to the control command receiving unit 400 of the gateway device 102 via the control command sending unit 305( Figure 6A and Figure 6B Step S608). That is, the learning contribution determination unit 304 is an example of a publishing unit that publishes a stop command for the parameter acquisition device 103 as the designated configuration. The control command receiving unit 400 receives the operation stop command sent in step S608. The parameter acquisition device control unit 401 performs control so that the operation of the parameter acquisition device 103 designated in step S607 is stopped in response to the operation stop command received by the control command receiving unit 400( Figure 6A Step S609).

[0063] [Example 2]

[0064] In Example 2, since the system configuration, hardware configuration, and block diagrams of each configuration are the same as those in Example 1Figures 1 to 5 is the same as that shown, so the description of the block diagrams of these configurations will be omitted, and reference may be made to Figures 1 to 5 .

[0065] Figure 7A is a sequence diagram according to Example 2 of the present invention. Figure 7B is a flowchart showing the processing flow in the learning contribution determination unit 304 according to Example 2 of the present invention. In Figure 7A , for the sake of promoting visibility, the control command transmission unit 305 and the control command reception unit 400 that do not perform processing other than transmission and reception processing are omitted.

[0066] The learning processing unit 303 obtains input data for calculating the learning contribution rate from the input data storage unit 302 ( Figure 7A step S700). The learning processing unit 303 uses the input data obtained in step S700 to perform learning processing ( Figure 6A step S701), and sends the learning result to the learning contribution determination unit 304 ( Figure 7A step S702).

[0067] The learning contribution determination unit 304 receives the learning result sent in step S702, obtains the learning contribution rate including the parameters of the input data from the received learning result, and sets the parameters with a low learning contribution rate as low contribution parameters ( Figure 7A and Figure 7B step S703). Based on the determination information of the low contribution parameters in step S703, the learning contribution determination unit 304 sends a relearning command excluding the low contribution parameters to the learning processing unit 303 ( Figure 7A and Figure 7B step S704). That is, the learning contribution determination unit 304 gives an instruction to perform relearning using the input data excluding the parameters designated as low contribution parameters.

[0068] The learning processing unit 303 receives the relearning command sent in step S704, and in response to the received relearning command, obtains learning input data excluding the low contribution parameters from the input data storage unit 302 ( Figure 7A step S705). The learning processing unit 303 uses the learning input data obtained in step S705 to perform learning processing ( Figure 7A step S706). The learning processing unit 303 stores the prediction model generated through the learning processing in step S706 in the prediction model storage unit 306, and replaces the prediction model during actual operation with this prediction model by automatic or manual operation.

[0069] The learning contribution determination unit 304 determines the parameter acquisition device 103 that acquires the parameters considered to be low contribution parameters ( Figure 7A and Figure 7B step S707). That is, the learning contribution determination unit 304 designates the parameter acquisition device 103 that acquires the parameters considered to be low contribution parameters in step S707.

[0070] In this example, the learning contribution determination unit 304 determines that it is possible to refer to a list of parameter acquisition devices 103 that operate constantly regardless of the learning contribution rate used for the learning process (hereinafter referred to as the constant operation parameter acquisition device list) among the parameter acquisition devices 103. The constant operation parameter acquisition device list is pre-stored in a storage device such as the RAM 201, ROM 202, and secondary storage device 205 of the parameter optimization system 101.

[0071] The learning contribution determination unit 304 determines whether the parameter acquisition device 103 that acquires the parameters considered to be low contribution parameters is included in the constant operation parameter acquisition device list ( Figure 7A and Figure 7B step S708). When the parameter acquisition device 103 that acquires the parameters considered to be low contribution parameters is not included in the constant operation parameter acquisition device list ( Figure 7B in step S708 is "no"), the process proceeds to Figure 7A and Figure 7B step S709. When the parameter acquisition device 103 that acquires the parameters considered to be low contribution parameters is included in the constant operation parameter acquisition device list ( Figure 7B in step S708 is "yes"), the process proceeds to Figure 7A and Figure 7B step S711. That is, the learning contribution determination unit 304 makes the process different according to whether the parameter acquisition device 103 that acquires the parameters considered to be low contribution parameters is included in the constant operation parameter acquisition device list.

[0072] The learning contribution determination unit 304, in Figure 7A and Figure 7B step S709, sends an operation stop command for the parameter acquisition device 103 specified in step S707 to the control command receiving unit 400 of the gateway device 102 via the control command sending unit 305. The control command receiving unit 400 receives the operation stop command sent in step S709. The parameter acquisition device control unit 401 performs control so that the operation of the parameter acquisition device 103 specified in step S707 is stopped in response to the operation stop command received by the control command receiving unit 400 ( Figure 7A step S710).

[0073] The learning contribution determination unit 304, in Figure 7A and Figure 7B in step S711, via the control command sending unit 305, sends a communication stop command for the parameter acquisition device 103 specified in step S707 to the control command receiving unit 400 of the gateway device 102. The control command receiving unit 400 receives the communication stop command sent in step S709. The parameter sending control unit 403 performs control such that, in response to the communication stop command received by the control command receiving unit 400, the transmission of the parameters acquired by the parameter acquisition device 103 specified in step S707 is stopped ( Figure 7A step S711).

[0074] [Example 3]

[0075] In Example 3, since the system configuration, hardware configuration, and block diagrams of each configuration are the same as those Figures 1 to 5 shown in Example 1, the description of the block diagrams of these configurations will be omitted, and reference can be made to Figures 1 to 5 .

[0076] Figure 8A is a sequence diagram according to Example 3 of the present invention. Figure 8B is a flowchart showing the processing flow in the learning contribution determination unit 304 according to Example 3 of the present invention. In Figure 8A , for the sake of promoting visibility, the configurations in which no processing other than the sending and receiving processing in this example is performed and the configurations in which only preprocessing of the collected data is performed will be omitted. In this example, it is assumed that the operations of some of the parameter acquisition devices 103 are pre-stopped by the functions described in Example 1 and Example 2.

[0077] In this example, the parameter acquisition device control unit 401 sends a temporary operation command to the parameter acquisition device 103 that has stopped operating (hereinafter referred to as the parameter acquisition device using the stopped one) at a predetermined timing ( Figure 8A steps S800 and S801). The predetermined timing at which the parameter acquisition device control unit 401 sends the temporary operation command can be a given periodic timing, can be an aperiodic timing, or can be a timing corresponding to the occurrence of a predetermined event. The parameter acquisition device control unit 401 can send the temporary operation command to the parameter acquisition device using the stopped one in response to a manual operation.

[0078] The parameter acquisition device using the stopped one that has received the temporary operation command operates to acquire parameters ( Figure 8A step S802). The gateway device 102 sends the parameters acquired by the parameter acquisition device 103 in step S802 to the parameter optimization system 101 ( Figure 8AStep S803). The gateway device 102 sends parameters via the parameter receiving unit 404, the parameter sending control unit 403, and the parameter sending unit 402.

[0079] The parameter optimization system 101 receives the parameters acquired by the parameter acquisition device 103 in step S802, and accumulates the received parameters in the input data storage unit 302 via the collected data receiving unit 300 and the input data generation unit 301. When the parameters acquired by the parameter acquisition device 103 in step S802 are sufficiently accumulated in the input data storage unit 302, the learning processing unit 303 obtains the input data for calculating the learning contribution rate from the input data storage unit 302 ( Figure 8A Step S804). When a predetermined condition is satisfied, it is assumed that the parameters are sufficiently accumulated in the input data storage unit 302. The predetermined condition is, for example, a condition that the number of input data accumulated after the operation of the stopped parameter acquisition device exceeds a threshold.

[0080] The learning processing unit 303 performs learning processing using the input data obtained in step S804 ( Figure 8A Step S805). The learning processing unit 303 sends the learning result learned in step S805 to the learning contribution determination unit 304 ( Figure 8A Step S806). The learning contribution determination unit 304 receives the learning result sent in step S806, obtains the learning contribution rate of the parameters included in the input data from the received learning result, and sets the parameters with a low learning contribution rate as low contribution parameters ( Figure 8A and Figure 8B Step S807).

[0081] The learning contribution determination unit 304 determines whether the learning contribution rate of the parameters acquired by the temporarily operated stopped parameter acquisition device (hereinafter referred to as stopped parameters) is low without changing from the learning contribution rate at the previous learning ( Figure 8A and Figure 8B Step S808). That is, in Figure 8B Step S808, even in the learning of step S805, the learning contribution determination unit 304 determines whether the stopped parameters are low contribution parameters. When the learning contribution determination unit 304 determines that the stopped parameters are low contribution parameters without changing from the parameters at the previous learning ( Figure 8B “Yes” in step S808), the process ends as it is. When the learning contribution determination unit 304 determines that the stopped parameters change from the parameters at the previous learning and are not low contribution parameters (they contribute sufficiently) ( Figure 8B “No” in step S808), the process proceeds to Figure 8A and Figure 8BStep S809. The learning contribution determination unit 304 sends, in Figure 8A and Figure 8B step S808, a relearning command for giving a relearning instruction including a parameter for use stop to the learning processing unit 303.

[0082] The learning processing unit 303 receives the relearning command sent in step S809, and in response to the received relearning command, obtains learning input data including a parameter for use stop from the input data storage unit 302 ( Figure 8A step S810). The learning processing unit 303 performs learning processing using the learning input data obtained in step S809 ( Figure 8A step S811). The learning processing unit 303 stores the prediction model generated through the learning processing in step S811 in the prediction model storage unit 306, and replaces the prediction model during actual operation with this prediction model by automatic or manual operation.

[0083] The learning contribution determination unit 304 determines the parameter acquisition device 103 that acquires a parameter for use stop that is not considered a low contribution parameter ( Figure 8A and Figure 8B step S812). That is, the learning contribution determination unit 304 specifies the parameter acquisition device 103 that acquires a parameter for use stop that is not considered a low contribution parameter in step S812. The learning contribution determination unit 304 sends an operation restoration command for the parameter acquisition device 103 specified in step S812 to the control command receiving unit 400 of the gateway device 102 via the control command sending unit 305 ( Figure 8A and Figure 8B step S813). The control command receiving unit 400 receives the operation restoration command sent in step S813. The parameter acquisition device control unit 401 performs control so that the operation of the parameter acquisition device 103 specified in step S812 is restored in response to the operation restoration command received by the control command receiving unit 400 ( Figure 8A step S814).

[0084] Other embodiments

[0085] Embodiments of the present invention can also be implemented by a computer of a system or apparatus that reads and executes computer-executable instructions (e.g., one or more programs) recorded on a storage medium (which may also be more fully referred to as a "non-transitory computer-readable storage medium") to perform one or more of the functions in the above embodiments, and / or includes one or more circuits (e.g., an application-specific integrated circuit (ASIC)) for performing one or more of the functions in the above embodiments. Moreover, embodiments of the present invention can be implemented by a method in which the computer of the system or apparatus, for example, reads and executes the computer-executable instructions from the storage medium to perform one or more of the functions in the above embodiments, and / or controls the one or more circuits to perform one or more of the functions in the above embodiments. The computer may include one or more processors (e.g., a central processing unit (CPU), a microprocessing unit (MPU)), and may include a network of separate computers or separate processors to read and execute the computer-executable instructions. The computer-executable instructions may be provided to the computer, for example, from a network or the storage medium. The storage medium may include, for example, a hard disk, a random access memory (RAM), a read-only memory (ROM), a memory of a distributed computing system, an optical disc (such as a compact disc (CD), a digital versatile disc (DVD), or a Blu-ray disc (BD) TM ), a flash device, and a memory card, among others.

[0086] Embodiments of the present invention can also be implemented by the following method, i.e., by providing software (a program) that performs the functions of the above embodiments to a system or apparatus via a network or various storage media, and the method in which a computer or a central processing unit (CPU), a microprocessing unit (MPU) of the system or apparatus reads and executes the program.

[0087] Although the present invention has been described with reference to exemplary embodiments, it should be understood that the present invention is not limited to the disclosed exemplary embodiments. The scope of the appended claims should be given the broadest interpretation so as to cover all such variations and equivalent structures and functions.

[0088] This application claims the benefit of Japanese Patent Application No. 2019-142936, filed on Aug. 2, 2019, the entire contents of which are incorporated herein by reference.

Claims

1. A system for managing a prediction model, the prediction model being generated by a learning process using input data including a plurality of parameters, the system comprising: a first calculation unit configured to calculate a learning contribution rate of each of the plurality of parameters according to a learning result; a first specifying unit configured to specify parameters with each learning contribution rate less than a threshold according to the learning contribution rates calculated by the first calculation unit, wherein the threshold is used to determine one or more configurations to be stopped; a first instruction unit configured to give a relearning instruction using input data excluding the parameters specified by the first specifying unit; a second specifying unit configured to specify a configuration corresponding to the parameters specified by the first specifying unit, wherein the configuration corresponds to at least one of a device, a sensor, and software; and a first issuing unit configured to issue a stop command for the configuration specified by the second specifying unit.

2. The system according to claim 1, wherein, the first issuing unit makes the processing different according to whether the configuration specified by the second specifying unit is included in a list.

3. The system according to claim 2, wherein, the first issuing unit issues an operation stop command for the configurations specified by the second specifying unit that are not included in the list, and issues a communication stop command for the configurations specified by the second specifying unit that are included in the list.

4. The system according to any one of claims 1 to 3, the system further comprising: an acquisition unit configured to temporarily acquire the parameters specified by the first specifying unit under a predetermined condition; a second instruction unit configured to give a relearning instruction using input data including the temporarily acquired parameters; a second calculation unit configured to calculate a learning contribution rate of the parameters specified by the first specifying unit according to a learning result of relearning using input data including the temporarily acquired parameters; a third specifying unit configured to specify a configuration corresponding to the parameters with a learning contribution rate not lower than that calculated by the second calculation unit; and a second issuing unit configured to issue a recovery command for the configuration specified by the third specifying unit.

5. The system according to claim 1, wherein, the prediction model predicts control of one of a smart home, an aircraft, a ship, a robot including a drone, an image processing device including at least one of a printing device and a scanner, a 3D printer, and a communication relay device.

6. A method performed by a system, the system managing a prediction model generated using input data including a plurality of parameters in a learning process, the method comprising: a first calculation step of calculating a learning contribution rate of each of the plurality of parameters according to a learning result; a first specifying step of specifying parameters with each learning contribution rate less than a threshold according to the learning contribution rates calculated in the first calculation step, wherein the threshold is used to determine one or more configurations to be stopped; The first instruction step, giving a relearning instruction for input data using parameters excluded from those specified in the first specifying step; The second specifying step, specifying a configuration corresponding to the parameters specified in the first specifying step, where the configuration corresponds to at least one of a device, a sensor, and software; and The first issuing step, issuing a stop command for the configuration specified in the second specifying step.

7. A storage medium storing a program that causes a computer to function as the respective units of the system according to claim 1.

Images