WO2023136191A1 - Dispositif de traitement d'informations, procédé de traitement d'informations et programme - Google Patents

Dispositif de traitement d'informations, procédé de traitement d'informations et programme Download PDF

Info

Publication number
WO2023136191A1
WO2023136191A1 PCT/JP2023/000050 JP2023000050W WO2023136191A1 WO 2023136191 A1 WO2023136191 A1 WO 2023136191A1 JP 2023000050 W JP2023000050 W JP 2023000050W WO 2023136191 A1 WO2023136191 A1 WO 2023136191A1
Authority
WO
WIPO (PCT)
Prior art keywords
inference
information
input information
data source
input
Prior art date
Application number
PCT/JP2023/000050
Other languages
English (en)
Japanese (ja)
Inventor
竹識 板垣
Original Assignee
ソニーグループ株式会社
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 ソニーグループ株式会社 filed Critical ソニーグループ株式会社
Publication of WO2023136191A1 publication Critical patent/WO2023136191A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F8/00—Arrangements for software engineering
    • G06F8/70—Software maintenance or management
    • G06F8/71—Version control; Configuration management
    • 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
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models

Definitions

  • the present technology relates to an information processing device, an information processing method, and a program, and more particularly to an information processing device, an information processing method, and a program that can prevent malfunction due to execution of inference based on unintended input information.
  • Japanese Patent Laid-Open No. 2002-200001 proposes a mechanism that, when a version mismatch between firmware and a driver is detected, upgrades/downgrades the versions to a correct combination so that the expected operation can be performed.
  • This technology has been developed in view of this situation, and is designed to prevent malfunctions caused by executing inferences based on unintended input information.
  • the information processing device or program of the present technology is a processing unit that executes inference based on input information, and when a mismatch occurs between the input information and input information assumed in advance, the inference or a program for causing a computer to function as such an information processing apparatus.
  • the processing unit of an information processing device having a processing unit executes inference based on input information, and when a mismatch occurs between the input information and presupposed input information Secondly, it is an information processing method for switching the operation mode of the inference.
  • inference is executed based on input information, and when a mismatch occurs between the input information and input information assumed in advance, the inference is performed.
  • the operation mode is switched.
  • FIG. 4 is a diagram for explaining a proper operating state of an electronic device assumed by a developer of an inference device; It is a figure explaining the actual operation condition. It is a figure showing an example of composition of an information processing system concerning a 1st embodiment of this art.
  • 2 is a block diagram showing an example of the internal configuration of an electronic device;
  • FIG. 10 is a flowchart illustrating the overall flow of inference processing; 6 is a flow chart showing an example procedure of data source abnormality detection processing according to the first embodiment;
  • FIG. 10 is a diagram showing an example of a version information list;
  • FIG. 11 is a diagram showing an example of the contents of a support version table; 7 is a flow chart showing an example of an inference trial process procedure according to the first embodiment;
  • FIG. 11 is a diagram illustrating an inference execution table; 6 is a flow chart showing an example of a procedure of device control processing according to the first embodiment; 6 is a flow chart showing an example of a procedure of abnormality notification processing according to the first embodiment; FIG. 11 is a flow chart showing an example of an inference trial process procedure according to the second embodiment; FIG. FIG. 11 is a flow chart showing an example of a procedure of abnormality notification processing according to the second embodiment; FIG. FIG. 11 is a flow chart showing an example procedure of data source abnormality detection processing according to the third embodiment; FIG. FIG. 2 is a block diagram showing a hardware configuration example of a computer that executes a series of processes by a program;
  • Today's intelligent equipment (called electronic equipment) consists of various sensors, communication devices, and reasoning devices (AI: artificial intelligence) is installed.
  • AI artificial intelligence
  • One or more feature values conforming to the specifications assumed by the developer are input to the inference machine as input information, and the inference results based on the specifications of the input information are output as output information from the inference machine. .
  • each component such as sensors and communication devices installed in electronic equipment may be updated independently.
  • Component updates include cases where hardware is updated (including changes), and cases where software such as firmware and driver software (hereinafter referred to as drivers) that operate hardware are updated (including changes).
  • Fig. 1 is a diagram explaining the proper operational status of an electronic device assumed by the developer of the reasoner.
  • the electronic device assumed by the developer has a reasoner, and as components that provide input information to the reasoner, for example, sensors (sensor devices) shown in the figure, other software modules, and communication devices.
  • Other software modules are software other than software (firmware and drivers) related to sensors and communication devices, and are executed by hardware including CPU (Central Processing Unit), memory, etc. that electronic devices have as components of the main body. It represents the software module that provides input information (feature values) to the inferrer, among multiple types of software modules that include the program to be executed.
  • Other software modules are hereinafter simply referred to as software modules.
  • Data such as measured values obtained by sensors, calculation results obtained by software modules, user operation history, and statistics obtained by communication devices, for example, are supplied to the reasoner as reference information.
  • the reference information is adapted to the input specifications of the reasoner and input to the reasoner as input information (feature quantity).
  • the reasoner makes predictions about predetermined content of inference based on input information, and outputs the suggestion result (inference result) as output information.
  • an electronic device performs communication (Internet communication) with an external device via the Internet.
  • the predetermined content of inference is the degree of confidence (" degree of risk of Internet communication delays”). Confidence is a value between 0 and 1.
  • Each component such as sensors, software modules, and communication devices has a revision stage, and a version indicating the revision stage is attached.
  • the versions of sensors, software modules, and communication devices when a developer develops a reasoner are version V.1.1.5, version V.1.0.6, and version V. .1.0.2.
  • the developer determines the input information (feature values) to be input to the inference machine and the output Determine the output information (inference content) to be used.
  • the reasoner performs arithmetic processing according to an inference model having, for example, a neural network (deep neural network) structure.
  • An inference model is learned in advance and parameters such as weights and biases of the inference model are determined before it is used as an inference device in an electronic device.
  • the inference device outputs appropriate output information (inference result) is output.
  • Fig. 2 is a diagram for explaining the actual operation situation with respect to Fig. 1.
  • the output of the updated component Specifications subject to change may differ from the reference information assumed by the reasoner developer.
  • the version of the software module is changed from version V.1.0.6 in FIG. 1 to version V.2.0.0 due to version upgrade.
  • Patent Document 1 Japanese Patent Application No. 2017-28930
  • the versions are expected to be upgraded/downgraded to a correct combination. It provides a mechanism to make it work.
  • this only guarantees that the device will work properly on its own, and it does not take into consideration the linkage with the inference machine that uses it, and the specifications are different from the assumptions of the inference machine developers. In some cases, the aforementioned problems persist.
  • This technology detects a state in which unintended (assumed) input information is input to the inference device, and changes the mode of inference operation (inference processing) according to the situation, thereby preventing malfunction of the electronic device. Provide a safe mechanism.
  • FIG. 3 is a diagram showing a configuration example of an information processing system according to the first embodiment of the present technology.
  • the information processing system 11 has an electronic device 21 and an information aggregation server 22 .
  • the electronic device 21 and the information aggregation server 22 are communicably connected via the Internet.
  • the communication line that connects the electronic device 21 and the information aggregation server 22 is assumed to be the Internet, but any communication line may be used.
  • the electronic device 21 is, for example, a terminal device such as a smartphone owned by a user, a PC (personal computer), a mobile PC, a tablet, etc., and corresponds to the electronic device shown in FIGS.
  • the electronic device 21 is not limited to a specific type, and may be any type of information processing device.
  • the electronic device 21 is assumed to have an inferring device that infers “the degree of risk of Internet communication being disrupted”.
  • the inference content of the inference device is not limited to this.
  • the information aggregation server 22 is a server device that can be accessed by the developer of the reasoner, and collects and provides information related to the reasoner installed in the electronic device 21 .
  • a reasoner developer can provide instruction information, including instructions regarding the operation of the reasoner, to the electronic device 21 via the information aggregation server 22 .
  • FIG. 4 is a block diagram showing an internal configuration example of the electronic device 21 of FIG. 4, the electronic device 21 has data sources 41-1 to 41-3 (data sources #1 to #3), an inference control section 42, a device control section 43, and a version database 44.
  • data sources 41-1 to 41-3 data sources #1 to #3
  • inference control section 42 an inference control section 42
  • device control section 43 a device control section 43
  • version database 44 a version database 44
  • the data sources 41-1 to 41-3 (#1 to #3) generate input information (input information for inference) to the inference device of the inference control unit 42 among the components of the electronic device 21. It represents a component that provides reference information for (hereinafter also referred to as reasoner reference information).
  • the electronic device 21 has three data sources 41-1 to 41-3 (#1 to #3), but the number of data sources is not limited to this.
  • the data sources 41-1 to 41-3 are represented by data sources #1 to #3 using data source numbers.
  • Data sources #1 to #3 are assumed to be a predetermined sensor device that is a sensor type, a predetermined communication device that is a communication device type, and a predetermined software module that is a software component, respectively.
  • the sensor device of data source #1 is, as a specific example, acceleration sensor hardware, firmware, and drivers.
  • the acceleration sensor hardware is the physical component of the acceleration sensor.
  • the firmware of the acceleration sensor is software incorporated in the hardware of the acceleration sensor and controlling the operation of the hardware.
  • the acceleration sensor driver is software related to the acceleration sensor, and is software executed by hardware (such as a CPU) on the main body side of the electronic device 21 .
  • the acceleration sensor driver performs, for example, hardware control via the acceleration sensor firmware, and calculation and output processing of acceleration measurement values based on signals provided from the acceleration sensor hardware and firmware. Note that the hardware, firmware, and driver of the acceleration sensor are simply referred to as the acceleration sensor.
  • the communication device of data source #2 is, as a specific example, the hardware, firmware, and driver of a wireless LAN (Local Area Network) device.
  • the hardware of a wireless LAN device is a physical component of the wireless LAN device, and the firmware of the wireless LAN device is software incorporated in the hardware of the wireless LAN device to control the operation of the hardware.
  • the wireless LAN device driver is software related to the wireless LAN device, and is software executed by hardware on the main body side of the electronic device 21 .
  • the driver of the wireless LAN device controls hardware via the firmware of the wireless LAN device, exchanges communication data with the wireless LAN device and processing units on the main body side of the electronic device 21, and controls communication connection. .
  • the hardware, firmware, and driver of a wireless LAN device are simply referred to as a wireless LAN device.
  • the software module of data source #3 is assumed to be a user context information provision module as a specific example.
  • the user context information providing module is software executed by hardware on the main body side of the electronic device 21 .
  • the user context information providing module refers to the past operation history and movement history of the electronic device 21, and calculates whether or not the wireless environment in which the electronic device 21 is currently placed is home.
  • the user context information providing module may be an independent application, a software module incorporated in any application, or a part of the operating system.
  • the inference control unit 42 performs preprocessing such as normalization on the reference information from the data sources #1 to #3, and then supplies input information (feature amounts) to the included inference units (inference units A to C). Enter as The reasoner predicts (estimates) the "degree of risk of Internet communication being disrupted” based on the input information, and outputs a value in the range of 0 to 1 corresponding to that degree as output information (inference result). do.
  • the inference control unit 42 supplies output information (inference result) of the inference unit to the equipment control unit 43 .
  • the inference control unit 42 can include one or more reasoners A to C.
  • the inference control unit 42 performs inference based on the versions of the data sources #1 to #3 in the current environment (current time) stored in the version database 44 and an inference operation table described later stored in the inference control unit 42. to determine the mode of inference operation (mode of inference operation). Determining the mode of the inference operation means determining which one of the inferencers A to C is used for inference, and also determining the type of input information (feature values) to be input to the inferencer.
  • the equipment control unit 43 controls the overall operation of the electronic equipment 21 based on the output information (inference result) of the inference unit supplied from the inference control unit 42 . From the inference control unit 42, the device control unit 43 is supplied with the "degree of risk of delay in Internet communication" as an inference result. In the description of the present technology, for example, when the value of the inference result is large (when it is equal to or greater than a predetermined threshold value), the device control unit 43 causes the display unit (UI: User Interface) (not shown) of the electronic device 21 to display “Internet There is a risk that communication will be disrupted.” will be output. Note that the notification of warning to the user is not limited to a specific method, and the processing of the device control unit 43 based on the result of inference is not limited to notification of warning.
  • the version database 44 aggregates and collectively manages the latest (current environment) versions of each of the data sources #1 to #3.
  • a version of each data source #1 to #3 is represented by a combination of versions of respective components of data sources #1 to #3.
  • the data source is a sensor or communication device, such as data source #1 or #2
  • the version of the data source is the version of each component hardware, firmware, and driver ( hardware version, firmware version, and device driver version).
  • the hardware version is not updated after production, and that the versions of data sources #1 and #2 are each represented by a combination of firmware version and device driver version.
  • the data source is a software component like data source #3, the version of the data source is represented only by the version (software version) of the software (software module) that is the software component.
  • the version database 44 has a mechanism for reading version information (version information) from each data source #1 to #3, or a mechanism for each data source #1 to #3 to write version information to the version database 44, and a version It has a storage unit that holds information.
  • the update of the version information is triggered by some change in built-in functions, such as when the electronic device 21 is started, when the software of the entire electronic device 21 is upgraded, or when a new application is installed. It is assumed that the version database 44 always stores and holds the latest version information.
  • FIG. 5 is a flowchart illustrating the overall flow of inference processing.
  • step S11 the inference control unit 42 performs data source abnormality detection processing.
  • the data source abnormality detection process it is determined whether or not the reference information of the reasoner supplied from the data sources #1 to #3 to the inference control unit 42 conforms to the specifications expected by the developer. done. That is, it is determined whether or not the input information from the data source used for inference is inconsistent with the presupposed input information. As a result, the presence or absence of an abnormality in the data source (data source abnormality) is detected.
  • the result of the data source abnormality detection process is obtained as a combination of two items: "presence or absence of data source abnormality" and "whether inference can be performed". Details of the data source abnormality detection process will be described later with reference to FIG.
  • step S12 the inference control unit 42 performs an inference trial process.
  • the inference trial process based on the result of the data source anomaly detection process in step S11, the aspect of the inference operation (including the case where no inference is performed) is determined, inference is performed accordingly, and the inference result (output from the inference unit information) to the device control unit 43 . Details of the inference trial process will be described later with reference to FIG.
  • step S13 the device control unit 43 performs device control processing.
  • a warning display or the like is performed based on the inference result of the inference trial process in step S12. Details of the device control processing will be described later with reference to FIG.
  • step S14 the inference control unit 42 performs abnormality notification processing.
  • the anomaly notification process information indicating that a data source anomaly has been detected in the data source anomaly detection process of step S11 is transmitted to the information aggregation server 22 and notified to the developer of the inference device. Details of the abnormality notification process will be described later with reference to FIG.
  • FIG. 6 is a flow chart showing an example of the procedure of data source abnormality detection processing.
  • step S31 when an inference request is triggered, the inference control unit 42 inquires of the version database 44 and reads the version information of each of the data sources #1 to #3 that provide reference information to the inference unit. . The process proceeds to step S32.
  • FIG. 7 is a diagram showing an example of a version information list (version information list) read from the version database 44.
  • the version information list records versions of firmware and drivers (firmware version and device driver version) of the acceleration sensor of data source #1 and the wireless LAN device of data source #2.
  • the software version is recorded in the user context information providing module of data source #3. If the data source has multiple components (hardware, firmware, and drivers) such as sensors and communication devices, uniquely identify the operation (specifications) of the data source among those components. Only versions of components that can be used may be recorded in the version information list. Version naming rules can vary from device to device.
  • the inference control unit 42 reads out the support version table held in advance by the inference control unit 42 itself.
  • FIG. 8 is a diagram showing an example of the contents of the support version table.
  • the support version table one or more combinations of versions of data sources #1 to #3 expected (assumed) by the developer are recorded as entries of the support version table.
  • An entry is a classification name of a combination of versions of data sources #1 to #3, and at the same time, represents a type (mode) of inference operation that can be performed by the inference control unit 42.
  • Different types have different entry numbers ( #1, #2, . . . ) are added. However, different entry numbers are allowed for the same type. Therefore, it means that the same type of inference operation is performed for a combination of versions of data sources #1 to #3 belonging to entries with the same entry number.
  • a combination of versions belonging to entries with five entry numbers are recorded as entries #1 to #5, but the number of entries is not limited to this.
  • Each entry #1 to #5 is classified as either "normal inference” or “fallback inference” as the operation type.
  • Normal inference is an inference operation in which input information (feature values) assumed (expected) by the developer is input to the inference machine based on reference information provided by data sources #1 to #3. means.
  • Each version of the data sources #1 to #3 that provide the reference information of the reasoner assumed by the developer is called a guaranteed version.
  • Fallback inference is when any of the data sources #1 to #3 is a non-guaranteed version, and the reference information of the reasoner assumed by the developer is not provided from the data sources #1 to #3 (the developer assumed It means an inference operation that executes inference with accuracy that does not result in a failure (malfunction) as an operation of the electronic device 21 even if the input information (feature amount) to the inference device cannot be obtained. Therefore, a version combination where all of data sources #1 to #3 are guaranteed versions belongs to the entry that performs normal inference. A combination of versions where any one or more of data sources #1 to #3 is not a guaranteed version belongs to the entry that performs fallback inference. However, if any one or more of the data sources #1 to #3 are not guaranteed versions, the combination of versions may not belong to any entry. Since it is not possible, an inference operation of not inferring is performed.
  • entries #1 and #2 are entries for executing normal inference. Indicates that it will be implemented.
  • Entries #3 to #5 are entries for which fallback inference is performed, and fallback inference is performed when the version combination of data sources #1 to #3 is any of entries #3 to #5. represents The description of "*" in the version information of entries #3 to #5 means a wild card, meaning any version other than the guaranteed version whose operation cannot be guaranteed.
  • Entries #3 to #5 of these fallback operations test in advance the case where the developer cannot use all the input information (features) to the reasoner satisfactorily, and then, from the viewpoint of inference accuracy, Recorded as an acceptable combination.
  • the inference control unit 42 collates the support version table of FIG. 8 with the versions of the data sources #1 to #3 of the current environment acquired from the version database 44 in step S31 of FIG. Collation is performed in accordance with the priority determined in advance by the developer. Entries #1 to #5 in the support version table of FIG. 8 have higher priority in ascending order of entry number.
  • step S33 of FIG. 6 the inference control unit 42 determines that, as a result of the collation in step S32, an entry matching the combination of versions of data sources #1 to #3 in the current environment exists in the support version table of FIG. Determine whether or not
  • step S33 If the answer is NO in step S33, the process proceeds to step S34, and the inference control unit 42 obtains the result of "data source error” and "inference impossible".
  • step S35 the inference control unit 42 determines whether or not the operation type of the entry that matches the combination of the versions of the data sources #1 to #3 in the current environment is "normal inference". That is, the inference control unit 42 determines whether the entry that matches the combination of the versions of the data sources #1 to #3 in the current environment is the entry #1 or entry #2 of "normal inference".
  • step S35 If the result in step S35 is NO, the process proceeds to step S36, and the inference control unit 42 obtains the result of "data source abnormal" and "inference possible”. If the result in step S35 is affirmative, the process proceeds to step S37, and the inference control unit 42 obtains a result of "no data source failure" and "inference possible”.
  • the inference control unit 42 passes the determination results obtained in steps S34, S36, or S37 and the entries corresponding to the current environment data sources #1 to #3 to the inference trial process in step S12 of FIG.
  • FIG. 9 is a flow chart showing an example of an inference trial process procedure.
  • step S51 the inference control unit 42 determines whether the determination result of the data source abnormality detection process in FIG. It is determined whether or not there is an error in the source” and “inference is possible”).
  • step S51 determines whether the determination result in the data source abnormality detection process is "inference impossible" or not perform inference (inference skip), and the processing of this flowchart ends.
  • step S53 the inference control unit 42 refers to the inference execution table held by the inference control unit 42 itself.
  • the inference control unit 42 controls the inference operation mode corresponding to the entry detected in the data source abnormality detection process in step S11 (the entry that matches the version combination of the data sources #1 to #3 in the current environment) in the inference execution table. , select (determine) the inference device to be used for inference and the input information (feature values) to be input to the inference device.
  • FIG. 10 is a diagram exemplifying an inference execution table.
  • the inference execution table shows the mode of inference operation corresponding to each entry #1 to #5 of the support version table in FIG.
  • the mode of the inference operation includes the inference device to be used for inference, the types of input information (feature values) to the inference device based on the reference information provided from each data source #1 to #3, the handling thereof, and the inference It shows a judgment threshold (described later) when used in a subsequent stage for the output information (inference output) from the device.
  • the feature amount based on the reference information provided from the acceleration sensor (referred to as the feature amount from the acceleration sensor; hereinafter the same) is three-axis acceleration information (X acceleration, Y acceleration, and Z acceleration).
  • the feature value from the wireless LAN device is information about connection as statistical information obtained from communication with the access point.
  • the information about connection specifically includes reception level, physical layer modulation rate, channel frequency, successful transmission packet count, transmission failure packet count, channel busy rate, and packet retention amount.
  • a standby time length or a transmission/reception time length may be used as a feature amount.
  • the feature amount from the user context information provision module is a numerical value of the home determination result indicating whether or not the environment in which the electronic device 21 currently exists is inside the home. It should be noted that all feature quantities are input as input information to the reasoner after being subjected to appropriate normalization (normalization according to normal distribution or logarithmic normal distribution).
  • each feature amount as input information to the inference unit is shown in three ways: a circle, a triangle, and a cross, corresponding to each of the entries #1 to #5. ing.
  • Features marked with circles are input to the reasoner as usual.
  • the feature values indicated by triangle marks are masked (fixed) to predetermined reserved values and input to the inference unit without using the feature values from data sources #1 to #3.
  • Features marked with a cross are excluded from the input information to the reasoner. Input to the reasoner after being masked with a reserved value indicated by a triangle mark.
  • the inference control unit 42 includes three inferencers A to C.
  • the inference execution table indicates the reasoner used for inference corresponding to each of the entries #1 to #5.
  • reasoner A is used for inference.
  • the reasoner A is an 11-dimensional input reasoner (a reasoner having 11 input nodes) to which all 11 feature quantities shown in FIG. 10 can be input as input information.
  • reasoner A is used (selected) in normal inference performance.
  • reasoner A is used (selected) in performing fallback inference.
  • the presence determination result which is the feature quantity from the user context information providing module, is masked with a reserved value and input to the reasoner A.
  • reasoner B is used for inference.
  • the reasoner B is an 8-dimensional input reasoner capable of inputting, as input information, 8 types of feature values from the guaranteed version data sources #2 and #3 out of all the 11 types of feature values shown in FIG. (a reasoner with 8 input nodes).
  • the accelerometer of data source #1 is the non-guaranteed version, so reasoner B receives three feature quantities (X acceleration, Y acceleration, and Z acceleration) from the accelerometer. does not have one input node.
  • reasoner B is used in the fallback inference implementation and the X, Y, and Z accelerations from the accelerometer are not input to reasoner B.
  • reasoner C is used for inference.
  • the inference unit C can input 10 types of feature values excluding the buffer retention amount, which is one of the feature values from the data source #2 of the guaranteed version, out of all the 11 types of feature values shown in FIG. It is a 10-dimensional input reasoner (a reasoner with 10 input nodes).
  • the wireless LAN device of data source #2 is a non-guaranteed version. does not have
  • the reasoner C can input six feature values excluding the buffer retention amount from the non-guaranteed version of the data source #2 as input information. information) is not input at all. For example, parameters such as reception level, which are officially supported as OS functions, may be guaranteed to operate separately. Acquisition of some of the feature values may be guaranteed.
  • reasoner B is used in performing fallback inference, and the feature quantity from the wireless LAN device of data source #2, the buffer retention amount, is not input to reasoner C. Also, the channel busy rate, which is a feature quantity from the wireless LAN device of data source #2, is masked with a reserved value and input to reasoner C.
  • FIG. 8 The types of entries #1 to #5 shown in FIG. 8 and the types of reasoners A to C corresponding to the entries #1 to #5 shown in FIG. The types are not limited to these examples.
  • a case where inference is executed using all types of predetermined input information is defined as a first operation mode, and inference is performed without using some types of input information among all types of input information.
  • a case of executing inference is defined as a second operation mode, and a case of executing inference by masking some types of input information among all types of input information to a predetermined value is defined as a third operation mode,
  • a fourth mode of operation is when inference is not executed.
  • the inference control unit 42 may include only one or a plurality of inferencers of the type used for inference in the first or third operation modes. It may include one or more of the types of reasoners used for reasoning in two modes of operation, or inferences in both a second mode of operation and a third mode of operation.
  • step S53 of FIG. 9 the inference control unit 42 refers to the inference execution table to determine the inference device to be used for inference and the input information (feature amount processing operation) to the inference device. , the process proceeds to step S54.
  • step S54 the inference control unit 42 generates the feature amount determined in step S53 based on the reference information provided from the data sources #1 to #3, and uses the generated feature amount as input information to determine in step S53. Input to the inference machine. As a result, the inference control unit 42 executes inference. As a result of the inference execution, the inference device outputs a value in the range of 0 to 1, which represents the "degree of risk of Internet communication being disrupted" as an inference result (inference output). The inference result output from the inference unit is supplied from the inference control unit 42 to the device control unit 43, and used in the device control process in step S13 of FIG.
  • the inference control unit 42 determines that the current environment corresponds to entry #3 of the support version table in FIG. 8, and is determined as "fallback inference possible" ("data source abnormal” and "inference possible") Get results. As a result, the inference control unit 42 makes an affirmative determination in step S51 in the inference trial process of FIG. 9, and then refers to the mode of inference operation corresponding to entry #3 in the inference operation table of FIG. As a result, the inference control unit 42 uses the inference device A to perform fallback inference.
  • the inference control unit 42 as the input information to the inference unit A of the 11-dimensional input, for the 10-dimensional feature amount out of the 11-dimensional (11 types) feature amount, the feature amount from the acceleration sensor of the data source #1 X acceleration, Y acceleration, and Z acceleration, which are quantities, and reception level, physical layer modulation rate, channel frequency, transmission success packet count, transmission failure packet count, channel, which are feature quantities from the wireless LAN device of data source #2. Input the busy rate and packet retention amount to the reasoner A.
  • the user context information provision module of data source #3 is a version not covered by security, so the reserved value ( In the form, the value after normalization is assumed to be 0), and the reserved value 0 is input to the reasoner A. This allows fallback inference to be performed.
  • the inference result (output information) is output from the inference unit as the "degree of risk of Internet communication stagnation" as a value in the range of 0 to 1.
  • the inference result output from the inference unit is supplied from the inference control unit 42 to the device control unit 54 and used in the device control process in step S13 of FIG.
  • the input information to be input to the estimator Fallback inferences such as entries #3 and #5 that mask part of the input information (features) to the estimator, and entries #4 and #5 that reduce the input information (features) to the estimator, reduce the reliability of the inference results. point disadvantage.
  • the value (inference output) output from the reasoner as an inference result is a specific judgment such as "the degree of risk of Internet communication being interrupted" ) is true, the greater the inference output, the higher the confidence in the inference result.
  • the reliability of the inference output from the estimator is determined as in fallback inference. is lower than that of normal inference, the reliability of the judgment result with respect to the truth of the judgment, which is the content of inference, can be made equal to that of normal inference by setting the decision threshold higher than that of normal inference.
  • Judgment threshold values recorded corresponding to entries #1 to #5 in the inference operation table of FIG. It represents a determination threshold for determining false, and is determined to a value that makes the reliability of the determination result equivalent for each of the entries #1 to #5.
  • the decision threshold is 0.5, but for entries #3 to #5, where fallback inference is performed, the reliability of the inference output is reduced.
  • the judgment thresholds have been raised to 0.6, 0.6, and 0.8, respectively.
  • a determination threshold corresponding to the entry that determines the mode of the inference operation is supplied and used to determine the truth or falseness of the determination, which is the content of the inference.
  • step S13 Equipment control process (step S13) according to the first embodiment>
  • the device control process according to the first embodiment in step S13 of FIG. 5 will be described. Note that if "inference skip" (step S52 in FIG. 9) is selected in the inference trial process (inference trial moat in FIG. 9) in step S12 in FIG.
  • FIG. 11 is a flow chart showing an example of a procedure for device control processing.
  • the device control unit 43 outputs an inference output from the inference control unit 42, which is a value in the range of 0 to 1 indicating the "degree of risk of delay in Internet communication", and the data source abnormality in step S11.
  • a decision threshold corresponding to the entry detected by the detection process is obtained from the inference control unit 42 .
  • the device control unit 43 determines whether or not the inference output from the inference control unit 42 exceeds the determination threshold.
  • step S71 If the answer in step S71 is NO, the processing of this flowchart ends. If the result in step S71 is affirmative, the process proceeds to step S72, and the device control unit 43 causes a display unit (UI: User Interface) (not shown) to display that "there is a risk that Internet communication will be delayed.” After step S71, the processing of this flowchart ends.
  • the determination threshold is a value adjusted and determined by the developer. Since the optimum decision threshold is selected according to the inference device and the feature amount used for inference, it is guaranteed that the risk of malfunction will not increase regardless of the mode of inference operation.
  • the abnormality notification process is a process in which the inference control unit 42 transmits anomaly notification information such as an anomaly occurrence status of the data source to the developer (the information aggregation server 22 in FIG. 3).
  • Abnormality notification information is transmitted to a server managed by a developer (developer management server), and the information aggregation server 22 in FIG. 2 corresponds to the developer management server.
  • the abnormality notification process may be transmitted to the information aggregation server 22 by the device control unit 43 instead of the inference control unit 42, and is not limited to the case where a specific processing unit performs the abnormality notification process.
  • FIG. 12 is a flow chart showing an example of the procedure for anomaly notification processing.
  • the inference control unit 42 determines whether or not the result of the data source abnormality detection process in step S11 of FIG. 5 (flowchart of FIG. 6) is "data source abnormality". It should be noted that cases where it is determined that "data source is abnormal” include cases where “fallback inference is possible” and cases where "inference is not possible”.
  • step S91 the processing of this flowchart ends. If the determination in step S91 is affirmative, the process proceeds to step S92.
  • step S92 the inference control unit 42 transmits the following contents to the information aggregation server 22 as abnormality notification information.
  • ⁇ Data source error ⁇ Data source type ⁇ Version information list for each data source ⁇ Information that can identify the product (model number, product name) ⁇ When fallback inference is performed, ⁇ Used reasoner ⁇ Used features (including information on the presence or absence of features masked with reserved values) ⁇ Inference output value
  • the electronic device 21 By transmitting the abnormality notification information as described above to the information aggregation server 42, the electronic device 21 as a whole can be A mechanism is provided that can avoid situations where unexpected behavior occurs.
  • the process of step S92 ends, the process of this flowchart ends.
  • the inference can be performed.
  • the mode of operation is switched to prevent the accuracy of the inference result from deteriorating. Therefore, the electronic device 21 that also uses the inference result is prevented from malfunctioning.
  • the configuration example of the information processing system 11 according to the first embodiment in FIG. 3 and the configuration example of the first embodiment in FIG. is common with the configuration example of the electronic device 21 according to the form of . Therefore, the configuration examples of the information processing system and the electronic device according to the second embodiment are as described with reference to FIGS. 3 and 4, and detailed description thereof will be omitted.
  • FIG. 13 is a flow chart showing an example of an inference trial process procedure according to the second embodiment.
  • the inference control unit 42 receives an explicit instruction regarding the inference operation from the information aggregation server 22 in advance before reflecting the result of the data source abnormality detection process in step S11 of FIG. Determines whether it is set to perform an action.
  • the content of "explicit instruction about inference operation" is the inference unit to be used for inference, the feature amount used as input information to the inference unit (including information on whether there is a feature amount to be masked in the reserved value), and the inference output. Contains information on the determination threshold of
  • step S111 If the answer in step S111 is NO, the process proceeds to step S112.
  • Steps S112 to S115 have the same processing content as steps S51 to S54 in FIG. 9, and refer to the inference execution table held by the inference control unit 42 itself to determine the mode of inference operation and perform inference. . A detailed description is omitted.
  • step S111 If the result in step S111 is affirmative, the process proceeds to step S116.
  • step S116 the inference control unit 42 determines whether or not the explicit instruction regarding the inference operation from the information aggregation server 22 is to execute inference. If the result in step S116 is NO, the process proceeds to step S117, and in step S117, the inference control unit 42 does not perform inference (inference skip), and the process of this flowchart ends.
  • step S116 determines (selects) the inference device to be used for inference by the inference device and the feature amount to be used as input information to the inference device according to instructions from the information aggregation server 22 . Processing proceeds from step S118 to step S119.
  • step S119 the feature amount determined in step S118 is generated based on the reference information from data sources #1 to #3, and the generated feature amount is input to the inference unit determined in step S118 as input information.
  • the inference control unit 42 executes inference according to the mode of inference operation specified by the developer (information aggregation server 22).
  • the inference device outputs a value in the range of 0 to 1 indicating the "degree of risk of Internet communication being disrupted" as an inference result (inference output).
  • the inference result output from the inference unit and the determination threshold specified by the information aggregation server 22 are supplied from the inference control unit 42 to the device control unit 43 and used in the device control process in step S13 of FIG.
  • FIG. 14 is a flow chart showing an example of the procedure of abnormality notification processing according to the second embodiment.
  • the inference control unit 42 determines whether or not the result of the data source abnormality detection process in step S11 of FIG. 5 (flow chart of FIG. 6) is "data source abnormality". It should be noted that cases where it is determined that "data source is abnormal” include cases where "fallback inference is possible" and cases where "inference is not possible”.
  • step S131 If the result in step S131 is NO, the process skips step S132 and proceeds to step S133. If the determination in step S131 is affirmative, the process proceeds to step S132.
  • step S132 the inference control unit 42 transmits the following contents to the information aggregation server 22 as abnormality notification information.
  • ⁇ Data source error ⁇ Data source type ⁇ Version information list for each data source #1 to #3 ⁇ Information that can identify the product (model number and product name) ⁇ When fallback inference is performed, ⁇ Used reasoner ⁇ Used feature amount (including information on the presence or absence of the feature amount masked to the reserved value) ⁇ Inference output value
  • step S132 By transmitting the abnormality notification information as described above to the information aggregation server 22, the electronic device 21 as a whole can A mechanism is provided that can avoid situations where unexpected behavior occurs. Processing proceeds from step S132 to step S133.
  • step S133 the inference control unit 42 inquires of the information aggregation server 22 about the following information, and obtains a response.
  • ⁇ Inference operation instructions for this electronic device ⁇ Latest reasoner list ⁇ Latest support version table (with or without)
  • step S133 The difference from the abnormality notification process according to the first embodiment in FIG. 12 is that an inquiry is made in this step S133. Processing proceeds from step S133 to step S134.
  • step S134 the inference control unit 42, based on the response from the information aggregation server 22 to the inquiry in step S133, updates the newly added inference unit data and the latest support version table as necessary. Download and save. Assume, for example, that the inference control unit 42 has added a new inference device to be newly added in the information aggregating server 22 as a result of referring to the response to the inquiry about the “latest inference device list”. In that case, the inference control unit 42 downloads the data of the newly added inference device (data for executing the processing of the inference device) from the information aggregation server 22 and adds the inference device. Similarly, when the inference controller held by the inference control unit 42 has been updated, the data of the updated inference device is similarly downloaded from the information aggregation server 22 to update the inference device. may
  • the inference control unit 42 has updated the support version table to be held by the inference control unit 42 in the information aggregation server 22 as a result of referring to the reply to the inquiry about the "latest support version table".
  • the inference control unit 42 downloads the data of the support version table from the information aggregation server 22 and updates the support version table of the inference control unit 42 to the latest one. Processing proceeds from step S134 to step S135.
  • step S135 the inference control unit 42 saves and sets inference operation instructions in the electronic device 21 as necessary based on the response from the information aggregation server 22 to the inquiry in step S133. For example, the inference control unit 42 determines how the electronic device 21 (the inference control unit 42) should operate when an abnormality in the data source is detected as a result of referring to the response to the inquiry "Inference operation instruction for this electronic device.” Assume that the information aggregation server 22 has given an instruction regarding the mode of the inference operation of . In that case, the inference control unit 42 downloads the inference operation instruction from the information aggregation server 22, stores it, and sets it.
  • the inference control unit 42 receives an explicit instruction on the inference operation from the information aggregation server 22 in step S111 of the flowchart (inference trial processing) of FIG. It is determined that the inference operation is set by the instruction. For example, if the combination of versions of the data sources #1 to #3 in the current environment is not in the support version table held by the inference control unit 42, determination based on the support version table alone will result in inference skipping. However, inference may be possible if confirmation by the developer side can be obtained. Therefore, in this embodiment, the inference control unit 42 executes inference in accordance with the inference operation instruction from the information aggregation server 22 in such a case. After step S135, the processing of this flowchart ends.
  • the response from the information aggregation server 22 to the inquiry from the electronic device 21 (the inference control unit 42) is not always returned immediately. There may be a time lag between the response from 22 to the electronic device 21 .
  • the inference can be performed.
  • the mode of operation is switched to prevent the accuracy of the inference result from deteriorating. Therefore, the electronic device 21 that also uses the inference result is prevented from malfunctioning.
  • the fact is communicated to the developer (information aggregation server 22). is updated to As a result, it becomes possible to respond more flexibly to version updates of each data source.
  • FIG. 3 configuration examples of the information processing system and the electronic device according to the third embodiment are shown in FIG. This configuration is common to the configuration example of the electronic device 21 according to the first and second embodiments. Therefore, the configuration examples of the information processing system and the electronic device according to the third embodiment are as described with reference to FIGS. 3 and 4, and detailed description thereof will be omitted.
  • the data source anomaly detection process and anomaly notification process according to the third embodiment are different from those of the second embodiment. Therefore, only data source abnormality detection processing and abnormality notification processing according to the third embodiment, which are different from the second embodiment, will be described.
  • FIG. 15 is a flow chart showing an example procedure of data source abnormality detection processing according to the third embodiment.
  • the inference control unit 42 continues monitoring the input systems from the data sources #1 to #3 to the inference unit for a certain period of time.
  • the monitoring of the input system from each data source #1 to #3 to the reasoner indicates any of the following aspects.
  • the data source includes dedicated hardware such as sensors and communication devices
  • one or more output signals (data) output from the hardware are sent to the inference unit. This is a mode of monitoring (detecting) an output signal that contributes to input information (feature quantity).
  • the second aspect is when the data source includes dedicated hardware and includes software such as firmware and drivers as in the first aspect, or when the data source includes dedicated hardware such as software modules. is included, and only software is included, one or more signals (data) processed by the software are monitored (detected) for signals that contribute to input information (feature amounts) to the reasoner .
  • a third aspect is when the inference control unit 42 includes a preprocessing unit that generates input information (feature amounts) to the inference unit from a signal (data) supplied as reference information supplied from a data source. , the preprocessing unit detects a signal generated in the process of generating input information (feature amount) to the reasoner or the generated feature amount.
  • the inference control unit 42 continuously detects values of one or a plurality of input series from the data sources #1 to #3 to the inference device according to these first to third modes for a certain period of time. The process proceeds to step S151 or step S152.
  • step S152 the inference control unit 42 compares the value of the input system detected in step S151 with the value of the expected value series (hereinafter referred to as expected value).
  • the expected value indicates the average value of the input sequence values assumed by the developer. Processing proceeds from step S152 to step S153.
  • step S153 the inference control unit 42 determines whether or not there is a data source (feature value) in which the values of the input series detected in step S151 steadily deviate from the expected values.
  • a data source feature value
  • the value of the input system steadily deviates from the expected value for example, the time when the difference (absolute value of the difference) between the value of the input system and the expected value exceeds a predetermined threshold is If it continues longer than a predetermined time, or the number of input system values detected during a certain period of time, the difference (absolute value of the difference) from the expected value is greater than a predetermined threshold
  • a case is shown where the number of input system values that increase is greater than a predetermined ratio.
  • a data source whose input strain value regularly deviates from its expected value is a data source that supplies a signal for that input strain when its value consistently deviates from its expected value. means.
  • step S153 If the result in step S153 is NO, the process proceeds to step S154, and the inference control unit 42 obtains a result of "no data source abnormality" and "inference possible". If the determination in step S153 is affirmative, the process proceeds to step S155.
  • step S155 the inference control unit 42 refers to the inference operation table of FIG. It is determined whether or not there is an entry to be masked in , or an entry to be excluded from the input information to the reasoner.
  • step S155 If the result in step S155 is NO, the process proceeds to step S156, and the inference control unit 42 obtains the result of "data source error” and "inference impossible”. If the result in step S155 is affirmative, the process proceeds to step S157, and the inference control unit 42 obtains a result of "data source abnormal" and "fallback inference possible”.
  • the "channel busy rate" is one of the input information (feature values) from the wireless LAN device to the estimator.
  • the inference control unit 42 refers to the inference operation table of FIG. 10, and selects entry #5 in which the "channel busy rate" acquired from the wireless LAN device is masked with a reserved value as the mode of inference operation. be.
  • the inference control unit 42 obtains the results of "data source anomaly detected” and “fallback inference possible" as data source anomaly detection results.
  • Step S14 A procedure example of the abnormality notification process according to the third embodiment is substantially the same as the flowchart of FIG. 14 showing an abnormality notification process procedure example in the second embodiment.
  • the inference control unit 42 adds the following information to the information aggregation server 22. ⁇ Values (raw sample values) detected from the input series of the data source where the anomaly was detected (the input series where regular deviations from the expected values were detected)
  • the inference can be performed.
  • the mode of operation is switched to prevent the accuracy of the inference result from deteriorating. Therefore, the electronic device 21 that also uses the inference result is prevented from malfunctioning.
  • the data source abnormality detection process according to the third embodiment can also be applied to the first embodiment, and instead of detecting data source abnormality based on the data source version, It may be detected by the value of the input system from the data source to the reasoner.
  • the device control processing in step S13 and the abnormality notification processing in step S14 may be performed in any order, and may be performed in parallel.
  • the sensor device that can be the data source in FIG. 4 is not limited to the acceleration sensor, and may be any sensor such as an imaging device, a microphone, a touch sensor, a temperature sensor, a humidity sensor, a geomagnetic sensor, and a millimeter wave radar. .
  • Communication devices that can be data sources in FIG. 4 are not limited to wireless LAN devices, but are : Infrared Data Association), UWB®, Zigbee®, WiGig®, etc., any communication device.
  • the external software module that can be the data source in FIG. 4 may include an inference device other than the inference devices A to C of the inference control unit 42, and output the result as reference information.
  • the hardware version is used as the version information of the data source. may be included.
  • the abnormality notification information may be transmitted to the information aggregation server 22 via the PC.
  • the version of the user context information providing module of data source #3 in FIG. 4 is part of the OS (operating system) and does not have version information by itself, the version of the user context information providing module is version can be substituted.
  • (9) In the data source anomaly detection process according to the third embodiment shown in the flowchart of FIG. may be provided with a plurality of expected value sequences used in anomaly detection for comparing .
  • the inference control unit 42 uses a plurality of expected value sequences together with additional information for classifying situations and use cases, and performs anomaly detection using the expected value sequences corresponding to the situations and use cases.
  • the inference control unit 42 may transmit values (raw sample values) detected from the input series of all data sources to the information aggregation server 22. .
  • a series of processes in the electronic device 21 described above can be executed by hardware or by software.
  • a program that constitutes the software is installed in the computer.
  • the computer includes, for example, a computer built into dedicated hardware and a general-purpose personal computer capable of executing various functions by installing various programs.
  • FIG. 16 is a block diagram showing a hardware configuration example of a computer that executes the series of processes described above by a program.
  • a CPU Central Processing Unit
  • ROM Read Only Memory
  • RAM Random Access Memory
  • An input/output interface 205 is further connected to the bus 204 .
  • An input unit 206 , an output unit 207 , a storage unit 208 , a communication unit 209 and a drive 210 are connected to the input/output interface 205 .
  • the input unit 206 consists of a keyboard, mouse, microphone, and the like.
  • the output unit 207 includes a display, a speaker, and the like.
  • the storage unit 208 is composed of a hard disk, a nonvolatile memory, or the like.
  • a communication unit 209 includes a network interface and the like.
  • a drive 210 drives a removable medium 211 such as a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory.
  • the CPU 201 loads, for example, a program stored in the storage unit 208 into the RAM 203 via the input/output interface 205 and the bus 204 and executes the above-described series of programs. is processed.
  • the program executed by the computer (CPU 201) can be provided by being recorded on removable media 211 such as package media, for example. Also, the program can be provided via a wired or wireless transmission medium such as a local area network, the Internet, or digital satellite broadcasting.
  • the program can be installed in the storage section 208 via the input/output interface 205 by loading the removable medium 211 into the drive 210 . Also, the program can be received by the communication unit 209 and installed in the storage unit 208 via a wired or wireless transmission medium. In addition, programs can be installed in the ROM 202 and the storage unit 208 in advance.
  • the program executed by the computer may be a program that is processed in chronological order according to the order described in this specification, or may be executed in parallel or at a necessary timing such as when a call is made. It may be a program in which processing is performed.
  • An information processing apparatus comprising: a processing unit that performs inference based on input information, the processing unit switching an operation mode of the inference when a mismatch occurs between the input information and presupposed input information. .
  • a first operation mode is a case where the inference is executed using all the types of the input information determined in advance, and without using some types of the input information among all the types of the input information.
  • a case where the inference is executed is defined as a second operation mode, and a case where the inference is executed by masking some types of the input information out of all the types of the input information with a predetermined value.
  • a third mode of operation, and a fourth mode of operation when the inference is not executed The processing unit executes the inference in the first operation mode when the inconsistency does not occur, and executes the inference in one of the second to fourth operation modes when the inconsistency occurs.
  • the information processing apparatus according to (1) above.
  • the processing unit includes a first reasoner used for the reasoning in the first or third operation mode, the reasoning in the second operation mode, or the second operation mode, The information processing apparatus according to (2) or (3), further including a second reasoner used for the reasoning in the third operation mode.
  • the information processing apparatus determines that the inconsistency has occurred when the version of the data source that provides the input information is different from one or more versions assumed in advance, any of (1) to (5). 1.
  • the information processing device according to claim 1.
  • the information processing apparatus wherein the data sources are sensors, communication devices, or software components.
  • the version of the data source is specified by a combination of a plurality of versions of hardware, firmware, and device drivers, which are components of the data source.
  • the processing unit determines whether the mismatch has occurred based on a comparison between a value of an input system through which a signal contributing to the input information flows from a data source providing the input information and a predetermined expected value.
  • the information processing apparatus according to any one of (1) to (8).
  • the processing unit predicts, as an inference result, a degree of certainty that a specific judgment is true based on the inference.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Security & Cryptography (AREA)
  • Stored Programmes (AREA)

Abstract

La présente technologie concerne un dispositif de traitement d'informations, un procédé de traitement d'informations et un programme qui permettent d'empêcher une opération erronée due à l'exécution d'une inférence à l'aide d'informations d'entrée non souhaitées. Une inférence est exécutée sur la base d'informations d'entrée et s'il existe une incohérence entre les informations d'entrée et les informations d'entrée prédites, le mode de fonctionnement d'inférence est commuté.
PCT/JP2023/000050 2022-01-17 2023-01-05 Dispositif de traitement d'informations, procédé de traitement d'informations et programme WO2023136191A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2022-004859 2022-01-17
JP2022004859 2022-01-17

Publications (1)

Publication Number Publication Date
WO2023136191A1 true WO2023136191A1 (fr) 2023-07-20

Family

ID=87279028

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2023/000050 WO2023136191A1 (fr) 2022-01-17 2023-01-05 Dispositif de traitement d'informations, procédé de traitement d'informations et programme

Country Status (1)

Country Link
WO (1) WO2023136191A1 (fr)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170330109A1 (en) * 2016-05-16 2017-11-16 Purepredictive, Inc. Predictive drift detection and correction
JP2021039612A (ja) * 2019-09-04 2021-03-11 株式会社Uacj 情報処理装置、情報処理方法、および情報処理プログラム
WO2021162043A1 (fr) * 2020-02-14 2021-08-19 住友重機械工業株式会社 Système de machine de moulage par injection et machine de moulage par injection

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170330109A1 (en) * 2016-05-16 2017-11-16 Purepredictive, Inc. Predictive drift detection and correction
JP2021039612A (ja) * 2019-09-04 2021-03-11 株式会社Uacj 情報処理装置、情報処理方法、および情報処理プログラム
WO2021162043A1 (fr) * 2020-02-14 2021-08-19 住友重機械工業株式会社 Système de machine de moulage par injection et machine de moulage par injection

Similar Documents

Publication Publication Date Title
KR102339239B1 (ko) 클라우드-디바이스 협업적 실시간 사용자 사용 및 성능 비정상 검출을 위한 시스템 및 방법
US11616707B2 (en) Anomaly detection in a network based on a key performance indicator prediction model
US11210587B2 (en) Distributed learning anomaly detector
US10733536B2 (en) Population-based learning with deep belief networks
US10921775B2 (en) Production system
AU2019395267A1 (en) Explainability-based adjustment of machine learning models
WO2020046261A1 (fr) Analyse pronostique systématique avec modèle causal dynamique
US11271793B2 (en) Self-adjusting control loop
CN111130940A (zh) 异常数据检测方法、装置及服务器
US20120323344A1 (en) Automatically adjustable industrial control configuration
CN114450700A (zh) 用于检测异常的方法和设备、相应的计算机程序产品和非暂时性计算机可读载体介质
US20220237075A1 (en) Automated detection and classification of dynamic service outages
US11860721B2 (en) Utilizing automatic labelling, prioritizing, and root cause analysis machine learning models and dependency graphs to determine recommendations for software products
US10719629B1 (en) Strategy based feature disablement
WO2021048311A1 (fr) Procédé et appareil de gestion de prédiction d'anomalies de réseau
US20200159205A1 (en) Cloud and edge manufacturing data processing system
AU2021218159B2 (en) Utilizing machine learning models to determine customer care actions for telecommunications network providers
WO2023136191A1 (fr) Dispositif de traitement d'informations, procédé de traitement d'informations et programme
WO2020261621A1 (fr) Système et procédé de surveillance, et programme
EP4248308A1 (fr) Mises à jour d'application hiérachisées
US20230412682A1 (en) Adjusting a network of sensor devices
WO2018210876A1 (fr) Procédé et système de génération et de transmission à distance d'un procédé de prédiction d'état de dispositif local
US20230418691A1 (en) Environment-based device condition indicator for prioritized device-cloud interactions
US20230004469A1 (en) Management action predictions
US20230278567A1 (en) Autonomous driving control apparatus and method thereof

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: 23740200

Country of ref document: EP

Kind code of ref document: A1