WO2025069149A1 - 情報処理装置、情報処理方法、及び記録媒体 - Google Patents

情報処理装置、情報処理方法、及び記録媒体 Download PDF

Info

Publication number
WO2025069149A1
WO2025069149A1 PCT/JP2023/034766 JP2023034766W WO2025069149A1 WO 2025069149 A1 WO2025069149 A1 WO 2025069149A1 JP 2023034766 W JP2023034766 W JP 2023034766W WO 2025069149 A1 WO2025069149 A1 WO 2025069149A1
Authority
WO
WIPO (PCT)
Prior art keywords
risk
information processing
index
sequence data
time
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/JP2023/034766
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
章記 海老原
大輝 宮川
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NEC Corp
Original Assignee
NEC Corp
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 NEC Corp filed Critical NEC Corp
Priority to JP2025548033A priority Critical patent/JPWO2025069149A1/ja
Priority to PCT/JP2023/034766 priority patent/WO2025069149A1/ja
Publication of WO2025069149A1 publication Critical patent/WO2025069149A1/ja
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate

Definitions

  • This disclosure relates to the technical fields of information processing devices, information processing methods, and recording media.
  • Patent Document 1 discloses a device that classifies sequence data into one of a number of predefined classes by sequentially acquiring and analyzing multiple elements contained in the sequence data.
  • This disclosure aims to improve the related technology described above.
  • One aspect of the information processing device disclosed herein comprises an acquisition means for sequentially acquiring a finite number of elements contained in sequence data, an index calculation means for calculating an index indicating which of a plurality of classes the sequence data belongs to each time an element is acquired, a classification means for classifying the sequence data into one of the plurality of classes by comparing the index with a threshold, a risk calculation means for calculating the risk at each time by calculating backwards from the final time when all of the elements are acquired a recurrence formula that calculates the risk of misclassifying the sequence data based on the index, and a threshold calculation means for calculating the threshold so as to minimize the risk.
  • One aspect of the information processing method disclosed herein is an information processing method for calculating the threshold value used by an information processing device that includes an acquisition means for sequentially acquiring a finite number of elements contained in sequence data, an index calculation means for calculating an index indicating which of multiple classes the sequence data belongs to each time an element is acquired, and a classification means for classifying the sequence data into one of the multiple classes by comparing the index with a threshold value, and the method calculates the risk at each time by calculating backwards from the final time when all of the elements are acquired using a recurrence formula that calculates the risk of misclassifying the sequence data based on the index, and calculates the threshold value so as to minimize the risk.
  • One aspect of the recording medium disclosed herein is an information processing method for calculating the threshold value used by an information processing device that includes an acquisition means for sequentially acquiring a finite number of elements included in sequence data, an index calculation means for calculating an index indicating which of multiple classes the sequence data belongs to each time the element is acquired, and a classification means for classifying the sequence data into one of the multiple classes by comparing the index with a threshold value, and the information processing method includes a computer program recorded thereon that causes a computer to execute the information processing method, which calculates the risk of misclassifying the sequence data based on the index by calculating backward from the final time when all of the elements are acquired to calculate the risk at each time, and calculates the threshold value so as to minimize the risk.
  • FIG. 2 is a block diagram showing a hardware configuration of a first information processing apparatus.
  • FIG. 2 is a block diagram showing a functional configuration of a first information processing apparatus.
  • 10 is a flowchart showing a flow of a classification operation in the first information processing apparatus.
  • 10 is a flowchart showing a flow of a threshold value calculation operation in the first information processing apparatus.
  • FIG. 2 is a block diagram showing a functional configuration of a second information processing apparatus.
  • 10 is a flowchart showing a flow of a threshold value calculation operation in the second information processing apparatus.
  • 13 is a graph showing a method of calculating a conditional expected value in the second information processing apparatus.
  • 13 is a graph showing a threshold calculation method in the second information processing apparatus.
  • 10 is a graph showing an example of a threshold value calculated by the second information processing apparatus.
  • FIG. 1 First Embodiment The first embodiment will be described with reference to FIGS. 1 to 4.
  • Fig. 1 is a block diagram showing the hardware configuration of the first information processing apparatus.
  • the first information processing device 10 includes a processor 11, a RAM (Random Access Memory) 12, a ROM (Read Only Memory) 13, and a storage device 14.
  • the information processing device 10 may further include an input device 15 and an output device 16.
  • the above-mentioned processor 11, RAM 12, ROM 13, storage device 14, input device 15, and output device 16 are each connected via a data bus 17.
  • the data bus 17 may be an interface other than a data bus (e.g., LAN, USB, etc.).
  • the processor 11 reads a computer program.
  • the processor 11 is configured to read a computer program stored in at least one of the RAM 12, the ROM 13, and the storage device 14.
  • the processor 11 may read a computer program stored in a computer-readable storage medium using a storage medium reading device (not shown).
  • the processor 11 may obtain (i.e., read) a computer program from a device (not shown) disposed outside the information processing device 10 via a network interface.
  • the processor 11 controls the RAM 12, the storage device 14, the input device 15, and the output device 16 by executing the computer program that the processor 11 reads.
  • a functional block that executes various processes for classifying sequence data is realized within the processor 11.
  • the processor 11 may function as a controller that executes each control in the information processing device 10.
  • the processor 11 may be configured as, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an FPGA (field-programmable gate array), a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), or a quantum processor.
  • the processor 11 may be configured as one of these, or may be configured to use multiple processors in parallel.
  • RAM 12 temporarily stores computer programs executed by processor 11.
  • RAM 12 temporarily stores data that processor 11 uses temporarily while processor 11 is executing a computer program.
  • RAM 12 may be, for example, a D-RAM (Dynamic Random Access Memory) or an SRAM (Static Random Access Memory). Also, other types of volatile memory may be used instead of RAM 12.
  • ROM 13 stores computer programs executed by processor 11. ROM 13 may also store other fixed data. ROM 13 may be, for example, a P-ROM (Programmable Read Only Memory) or an EPROM (Erasable Read Only Memory). Also, other types of non-volatile memory may be used instead of ROM 13.
  • the storage device 14 stores data that the information processing device 10 stores long-term.
  • the storage device 14 may operate as a temporary storage device for the processor 11.
  • the storage device 14 may include, for example, at least one of a hard disk device, a magneto-optical disk device, an SSD (Solid State Drive), and a disk array device.
  • the input device 15 is a device that receives input instructions from a user of the information processing device 10.
  • the input device 15 may include, for example, at least one of a keyboard, a mouse, and a touch panel.
  • the input device 15 may be configured as a mobile terminal such as a smartphone or a tablet.
  • the input device 15 may be, for example, a device that includes a microphone and is capable of voice input.
  • the output device 16 is a device that outputs information related to the information processing device 10 to the outside.
  • the output device 16 may be a display device (e.g., a display) that can display information related to the information processing device 10.
  • the output device 16 may also be a speaker or the like that can output information related to the information processing device 10 as audio.
  • the output device 16 may be configured as a mobile terminal such as a smartphone or a tablet.
  • FIG. 1 shows an example of an information processing device 10 that is configured to include multiple devices, all or some of these functions may be realized by a single device.
  • Such an information processing device may, for example, be configured to include only the above-mentioned processor 11, RAM 12, and ROM 13, and the other components (i.e., storage device 14, input device 15, output device 16, etc.) may be provided by, for example, an external device connected to the information processing device 10.
  • the information processing device 10 may have some of its calculation functions realized by an external device (for example, an external server or cloud, etc.).
  • Fig. 2 is a block diagram showing the functional configuration of the first information processing device.
  • the first information processing device 10 is configured as a device for classifying time series data.
  • the first information processing device 10 may be configured as a device that acquires images of time series data and classifies the type of object contained in the image.
  • the first information processing device 10 is configured to include an acquisition unit 50, an index calculation unit 100, a classification unit 150, a risk calculation unit 210, and a threshold calculation unit 220 as components for realizing its functions.
  • Each of the acquisition unit 50, the index calculation unit 100, the classification unit 150, the risk calculation unit 210, and the threshold calculation unit 220 may be a processing block realized by, for example, the above-mentioned processor 11 (see FIG. 1).
  • the acquisition unit 50 is configured to be able to acquire sequence data.
  • the sequence data here refers to data including a number of elements arranged in a specific order, and one example is time-series data. More specific examples of sequence data include, but are not limited to, video data, audio data, or subdivided image data.
  • the acquisition unit 50 is configured to be able to sequentially acquire a finite number of elements included in the sequence data. For example, the acquisition unit 50 may be configured to acquire elements included in the sequence data one by one in order.
  • the acquisition unit 50 may acquire data directly from any data acquisition device (e.g., a camera, a microphone, etc.), or may read data that has been acquired in advance by a data acquisition device and stored in storage, etc. When acquiring data from a camera, the acquisition unit 50 may be configured to acquire data from each of a number of cameras.
  • the index calculation unit 100 is configured to be able to calculate an index from the sequence data acquired by the acquisition unit 50.
  • the "index" here is a value indicating to which of multiple classes that are candidates for classification of the sequence data the data belongs. More specific examples of the index will be described in other embodiments described later.
  • the index calculation unit 100 calculates an index each time an element is acquired. That is, the index calculation unit 100 calculates an index when the acquisition unit 50 acquires one element, and calculates a new index when the acquisition unit 50 acquires another element again.
  • the index calculation unit 100 may calculate an index using one element acquired immediately before. At this time, the index calculation unit 100 may calculate a new index using not only the acquired element but also indices calculated in the past (specifically, indices calculated from elements acquired in the past).
  • the classification unit 150 classifies the sequence data into one of a plurality of classes based on the index calculated by the index calculation unit 100. Specifically, the classification unit 150 classifies the sequence data into one of a plurality of classes by comparing the index calculated by the index calculation unit 100 with a preset threshold. For example, the classification unit 150 classifies the sequence data into class A when the index exceeds a threshold corresponding to class A. Furthermore, the classification unit 150 classifies the sequence data into class B when the index exceeds a threshold corresponding to class B.
  • the classification unit 150 may classify whether the face of a person in an image is a real face or a fake face (for example, a spoofed face using a photograph or a 3D mask). Furthermore, the plurality of classes may be three or more. In this case, the classification unit 150 may perform classification using a threshold set for each class. The threshold used by the classification unit 150 is calculated by the threshold calculation unit 220 described later.
  • the risk calculation unit 210 is configured to be able to calculate the risk when the classification unit 150 erroneously classifies sequence data. For example, the risk calculation unit 210 calculates a value indicating the risk when sequence data that should be classified as class A is classified as class B. The risk calculation unit 210 calculates the risk based on the index calculated by the index calculation unit 100. However, the risk calculation unit 210 may calculate the risk based on a true index (in other words, a correct label) that has been assigned to the sequence data in advance. The risk calculation unit 210 calculates the risk at each time an element included in the sequence data is acquired. Note that in this embodiment, "time" advances each time an element is acquired. For this reason, the time corresponds to the number of elements acquired so far.
  • the risk calculation unit 210 calculates the risk for each time by back-calculating a recurrence formula that calculates the risk based on an index from the final time at which all of the finite elements contained in the sequence data are acquired. Specifically, the risk calculation unit 210 first calculates the risk for the final time based on an index calculated when all of the elements have been acquired. From there, the risk calculation unit 210 calculates the risk for the time before the most recently acquired element is acquired by back-calculating using the recurrence formula. The risk calculation unit 210 calculates the risk for each time before the final time by repeating this back-calculation.
  • the recurrence formula may include a term corresponding to the penalty for classifying into the wrong class.
  • An example of a specific recurrence formula is the following formula (1).
  • R is risk
  • X (1,t) is sequence data up to the current time t.
  • K is the number of classes, and y is the class label.
  • p(y i
  • X (1,t) ) is the posterior probability of the class, and is an example of an index calculated by the index calculation unit 100.
  • ct is the sampling cost, which is a constant c multiplied by the current time t.
  • the sampling cost may be a function f(t) of the current time t.
  • the threshold calculation unit 220 is configured to be able to calculate a threshold used by the classification unit 150 based on the risk calculated by the risk calculation unit 210. Specifically, the threshold calculation unit 220 calculates a threshold so as to minimize the risk calculated by the risk calculation unit 210. The threshold calculation unit 220 calculates a threshold corresponding to the risk at each time. The threshold calculation unit 220 may calculate the thresholds in order from the final time in accordance with the risk calculation unit 210 calculating the risk by counting backwards from the final time. The threshold calculation unit 220 may be configured to calculate thresholds corresponding to all times, and then store the thresholds connected in chronological order as the final threshold.
  • a classification operation i.e., an operation of classifying sequence data into one of a plurality of classes
  • Fig. 3 is a flowchart showing the flow of the classification operation in the first information processing device.
  • the acquisition unit 50 first acquires one element included in the sequence data (step S101). Then, the index calculation unit 100 calculates an index based on the element acquired by the acquisition unit 50 (step S102).
  • the classification unit 150 determines whether there is a class in which the index calculated by the index calculation unit 100 exceeds a threshold (step S103). That is, the classification unit 150 compares the index calculated by the index calculation unit 100 with the threshold corresponding to each class, and determines whether the index exceeds the threshold corresponding to any class.
  • step S103 If there is a class whose index exceeds the threshold (step S103: YES), the classification unit 150 classifies the sequence data into a class whose index exceeds the threshold (step S104). On the other hand, if there is no class whose index exceeds the threshold (step S103: NO), the process is executed again from step S101. That is, the acquisition unit 50 acquires the next element contained in the sequence data, and repeats the same process as described above.
  • the index calculated by the index calculation unit 100 tends to approach the threshold corresponding to the class to be classified each time an element is acquired. Therefore, by repeating the process while acquiring elements one by one as described above, the sequence data can be classified into an appropriate class. At this time, for sequence data that is easy to classify, the threshold is exceeded at an early stage, so the sequence data can be classified early. On the other hand, for sequence data that is difficult to classify, elements continue to be acquired until the threshold is exceeded, so the sequence data can be classified accurately over a period of time.
  • a threshold calculation operation i.e., an operation of calculating a threshold used for classifying sequence data
  • a threshold calculation operation i.e., an operation of calculating a threshold used for classifying sequence data
  • the acquisition unit 50 first acquires sequence data (step S151). Unlike the classification operation described above, the acquisition unit 50 may acquire all elements contained in the sequence data together.
  • the sequence data acquired in the threshold calculation operation may be sequence data prepared for calculating the threshold (e.g., learning data with a correct answer label attached).
  • the index calculation unit 100 calculates an index for each time based on each element of the sequence data acquired by the acquisition unit 50 (step S152). That is, the index is calculated for each time when the elements included in the sequence data are acquired sequentially.
  • the risk calculation unit 210 calculates the risk based on the index calculated by the index calculation unit 100 (step S153). As already explained, the risk calculation unit 210 first calculates the risk at the final time. Then, the threshold calculation unit 220 calculates a threshold so as to minimize the risk calculated by the risk calculation unit 210 (step S154). The threshold calculation unit 220 calculates a threshold for the time corresponding to the risk calculated by the risk calculation unit 210. For example, when the risk calculation unit 210 calculates the risk at the final time, the threshold calculation unit 220 calculates the threshold for the final time.
  • the threshold calculation unit 220 determines whether or not to end the calculation of the threshold (step S155).
  • the threshold calculation unit 220 may determine to end the calculation of the threshold when, for example, the threshold has been calculated for all times (specifically, when the calculation backward from the final time to the first time has been completed).
  • the threshold calculation operation ends.
  • the threshold calculation unit 220 may determine not to end the calculation of the threshold, for example, if the threshold has not been calculated for all times. If it determines not to end the calculation (step S155: NO), it returns to the previous time (step S156) and repeats the process from step S153. By repeating the process in this manner, it continues to calculate backwards from the final time, and ultimately it is possible to calculate the risks and thresholds for all times.
  • the threshold value used to classify sequence data is calculated according to the risk calculated based on the index. In this way, it is possible to calculate a threshold value that can reliably reduce the risk when classifying. If the threshold value calculated in this way is used, more appropriate classification can be performed than, for example, when the threshold value is fixed at the same value from the beginning. Furthermore, more appropriate classification can be performed compared to when the threshold value is simply changed (for example, when it is monotonically decreased).
  • the second embodiment will be described with reference to Figures 5 to 9.
  • the second embodiment differs from the first embodiment described above only in some configurations and operations, and other parts may be similar to the first embodiment. Therefore, hereinafter, parts that differ from the first embodiment will be described in detail, and descriptions of other overlapping parts will be omitted as appropriate.
  • Fig. 5 is a block diagram showing the functional configuration of the second information processing device.
  • the same reference numerals are used to designate the same elements as those shown in Fig. 2.
  • the second information processing device 10 is configured to include, as components for realizing its functions, an acquisition unit 50, an index calculation unit 100, a classification unit 150, a risk calculation unit 210, and a threshold calculation unit 220.
  • the index calculation unit 100 in the second information processing device 10 includes a likelihood ratio calculation unit 101 and a posterior probability conversion unit 102.
  • the risk calculation unit 210 in the second information processing device includes a continuation risk calculation unit 211, a termination risk calculation unit 212, and a minimum risk holding unit 213.
  • the likelihood ratio calculation unit 101 is configured to be able to calculate a likelihood ratio based on elements included in the sequence data acquired by the acquisition unit 50.
  • the "likelihood ratio" here is an index indicating the likelihood of the class to which the sequence data belongs.
  • the likelihood ratio calculation unit 101 may calculate a likelihood ratio based on two consecutive elements among the elements sequentially acquired by the acquisition unit 50.
  • the likelihood ratio calculation unit 101 may calculate a likelihood ratio using a newly acquired element and a previously acquired element or a previously calculated likelihood ratio.
  • the likelihood ratio calculation unit 101 may include a storage unit that stores previously acquired elements or previously calculated likelihood ratios.
  • the likelihood ratio calculation unit 101 may be configured, for example, by a trained neural network.
  • the posterior probability conversion unit 102 is configured to be able to convert the likelihood ratio calculated by the likelihood ratio calculation unit 101 into a posterior probability.
  • the posterior probability is a value indicating the probability that an input element is classified into a specific class, and is a value that corresponds one-to-one to the likelihood ratio calculated by the likelihood ratio calculation unit 101.
  • the posterior probability is an index calculated by the index calculation unit 100, and is used for classification in the classification unit 150 and for risk calculation in the risk calculation unit 210.
  • the likelihood ratio calculated by the likelihood ratio calculation unit 101 may be used as an index.
  • the second information processing device 10 may be configured without including the posterior probability conversion unit 102. Whether the posterior probability or the likelihood ratio is used as an index, it is possible to appropriately calculate risk.
  • the continuing risk calculation unit 211 is configured to be able to calculate the continuing risk using the posterior probability converted by the posterior probability conversion unit 102. Alternatively, the continuing risk calculation unit 211 may be configured to be able to calculate the continuing risk using the likelihood ratio calculated by the likelihood ratio calculation unit 101.
  • the continuing risk is the risk when the next element is acquired by advancing the time without classifying the sequence data at the current time.
  • the continuing risk can also be said to be the risk expectation value at the next time conditioned by the likelihood ratio or posterior probability at the current time.
  • the continuing risk G ⁇ t can be calculated, for example, as shown in the following formula (2).
  • ⁇ is a vector that collects the posterior probabilities of all classes.
  • G is the minimum risk.
  • the minimum risk G will be described in detail in the explanation of the minimum risk holding unit 213 below.
  • the right side of the above formula (2) indicates the expected risk value when proceeding to time t+1 when the posterior probabilities up to time t are available.
  • the termination risk calculation unit 212 is configured to be able to calculate the termination risk using the posterior probability converted by the posterior probability conversion unit 102. Alternatively, the termination risk calculation unit 212 may be configured to be able to calculate the termination risk using the likelihood ratio calculated by the likelihood ratio calculation unit 101.
  • the termination risk is the risk when classifying sequence data at the current time.
  • the termination risk G st is a function of the likelihood ratio or posterior probability at the current time, and can be calculated, for example, as shown in the following formula (3).
  • the right side of the above formula (3) means to select j that minimizes the penalty in formula (1) shown in the first embodiment, and is the risk when completing the classification with the posterior probability currently at hand. Note that ⁇ i is the posterior probability of class i.
  • the minimum risk holding unit 213 is configured to be able to hold the smaller of the continuation risk calculated by the continuation risk calculation unit 211 and the termination risk calculated by the termination risk calculation unit 212 as the minimum risk.
  • the minimum risk is used to calculate the continuation risk at the previous time.
  • the termination risk can be held as the minimum risk as is. In this way, it becomes possible to calculate the continuation risk at the previous time by inverse calculation from the termination risk at the final time.
  • Fig. 6 is a flowchart showing the flow of the threshold calculation operation in the second information processing device.
  • Fig. 7 is a graph showing a calculation method of a conditional expected value in the second information processing device.
  • Fig. 8 is a graph showing a threshold calculation method in the second information processing device. Note that in Fig. 6, the same reference numerals are used for the same processes as those shown in Fig. 4.
  • step S251 when the threshold calculation operation in the second information processing device 10 is started, first the acquisition unit 50 acquires sequence data (step S251). Then, the likelihood ratio calculation unit 101 calculates the likelihood ratio for each time based on each element of the sequence data acquired by the acquisition unit 50 (step S252). Furthermore, the posterior probability conversion unit 102 converts the likelihood ratio calculated by the likelihood ratio calculation unit into a posterior probability (step S253). Note that when the likelihood ratio is used as an index, the processing of step S253 may be omitted.
  • the termination risk calculation unit 212 calculates the termination risk at the final time T (step S254).
  • the termination risk at the final time T calculated here is held by the minimum risk holding unit 213 as the minimum risk at the final time T.
  • the termination risk calculation unit 212 calculates the termination risk at the current time t (step S256). Note that the termination risk can be calculated at any time if the posterior probability or likelihood ratio is available. For this reason, the termination risk may be calculated at a different timing. For example, when determining the termination risk at the final time in step S254, the termination risk calculation unit 212 may also calculate the termination risks at other times together.
  • the continuing risk calculation unit 211 calculates the continuing risk at the current time t (step S257).
  • the continuing risk calculation unit 211 calculates the continuing risk using the minimum risk held by the minimum risk holding unit 213 (i.e., the minimum risk at the next time).
  • the conditional expectation is calculated using Gaussian process regression.
  • the Gaussian process is a type of generative model that generates an output from an input.
  • the input here is the posterior probability or likelihood ratio at time t.
  • the output is the minimum risk at time t+1.
  • the output points in the Gaussian process follow a Gaussian distribution, and the mean and variance are output.
  • the mean here becomes the conditional expectation.
  • the posterior probability ⁇ at the current time t is conditioned (i.e., if the graph is cut vertically at ⁇ t)
  • the cut also becomes a Gaussian distribution.
  • the average value ⁇ of the Gaussian distribution at the cut becomes the minimum risk at time t+1.
  • using Gaussian process regression makes it possible to easily calculate the conditional expectation (in other words, the continuing risk).
  • methods other than Gaussian process regression may be used to calculate the conditional expectation.
  • a joint probability distribution may be obtained and calculated using a model such as a normalized flow or a mixed Gaussian distribution.
  • the minimum risk holding unit 213 holds the smaller of the termination risk and the continuation risk as the minimum risk at the current time t (step S258).
  • the minimum risk held here will be used when calculating the continuation risk at the previous time.
  • the threshold calculation unit 220 stores the intersection of the termination risk and the continuation risk as a threshold (step S259).
  • the method of calculating the threshold will be explained below with reference to FIG. 8.
  • the threshold calculation unit 220 holds these intersections as thresholds. After calculating the thresholds for all times, the threshold calculation unit 220 joins them together in the time direction to arrive at the final threshold. In this way, it is possible to appropriately calculate a threshold that minimizes risk.
  • the threshold calculation unit 220 determines whether or not to end the calculation of the threshold (step S260). If it is determined that the calculation is to end (step S260: YES), the threshold calculation operation ends. On the other hand, if it is determined that the calculation is not to end (step S260: NO), the process starts again from step S255. That is, it returns to the previous time and repeats the same process as described above. By repeating the process in this way, the calculation proceeds backwards from the final time, and ultimately the risks and thresholds for all times can be calculated.
  • the second information processing device 10 calculates a threshold value that changes over time.
  • a threshold value that allows for early classification while reducing the risk of classification.
  • the series data A, B, and C shown in FIG. 9 have different trends in the fluctuation of the likelihood ratio, and while the series data A is relatively easy to classify, the series data C is relatively difficult to classify.
  • the threshold value is fixed at the initial value, classification of series data with a small gradient of the likelihood ratio will be delayed.
  • the threshold value changes over time, so all series data can be classified early and accurately.
  • the threshold value is changed to minimize the risk, allowing for early classification while minimizing the possibility of misclassification.
  • each embodiment also includes a processing method in which a program that operates the configuration of each embodiment to realize the functions of the above-mentioned embodiments is recorded on a recording medium, the program recorded on the recording medium is read as code, and executed on a computer.
  • computer-readable recording media are also included in the scope of each embodiment.
  • each embodiment includes not only the recording medium on which the above-mentioned program is recorded, but also the program itself.
  • the recording medium may be, for example, a floppy disk, hard disk, optical disk, magneto-optical disk, CD-ROM, magnetic tape, non-volatile memory card, or ROM.
  • the scope of each embodiment is not limited to programs recorded on the recording medium that execute processes by themselves, but also includes programs that operate on an OS in conjunction with other software or functions of an expansion board to execute processes.
  • the program itself may be stored on a server, and part or all of the program may be made downloadable from the server to a user terminal.
  • the program may be provided to the user in, for example, a SaaS (Software as a Service) format.
  • the information processing device described in Supplementary Note 1 includes an acquisition means for sequentially acquiring a finite number of elements included in sequence data, an index calculation means for calculating an index indicating to which of a plurality of classes the sequence data belongs each time an element is acquired, a classification means for classifying the sequence data into one of the plurality of classes by comparing the index with a threshold, a risk calculation means for calculating the risk at each time by calculating backwards from the final time at which all of the elements are acquired a recurrence formula that calculates a risk of misclassifying the sequence data based on the index, and a threshold calculation means for calculating the threshold so as to minimize the risk.
  • the information processing device described in Supplementary Note 2 is the information processing device described in Supplementary Note 1, wherein the risk calculation means calculates a continuation risk, which is the risk when the next element is obtained without classifying sequence data at the current time, and a termination risk, which is the risk when classifying sequence data at the current time, and the threshold calculation means calculates an intersection of the continuation risk and the termination risk as the threshold.
  • Appendix 3 The information processing device described in Appendix 3 is the information processing device described in Appendix 3, wherein the risk calculation means holds the smaller of the continuing risk and the termination risk at the current time as the minimum risk at the current time, and calculates the continuing risk at the time one time before the current time using the minimum risk at the current time.
  • Appendix 4 The information processing device described in Appendix 4 is the information processing device described in Appendix 2, wherein the risk calculation means sets the value of the minimum risk at the final time as the value of the termination risk at the final time, and then calculates the continuing risk at each time by calculating backwards from the final time a recurrence formula for determining the continuing risk.
  • the information processing device according to Supplementary Note 5 is the information processing device according to any one of Supplementary Notes 2 to 4, wherein the risk calculation means calculates the continuing risk using Gaussian process regression.
  • the information processing device described in Supplementary Note 6 is the information processing device described in any one of claims 1 to 5, wherein the index is a likelihood ratio indicating the likelihood that the sequence data belongs to a certain class of the multiple classes, or a posterior probability corresponding to the likelihood ratio.
  • the information processing method described in Supplementary Note 7 is an information processing method for calculating the threshold used by an information processing device including: acquisition means for sequentially acquiring a finite number of elements included in sequence data; index calculation means for calculating an index indicating to which of a plurality of classes the sequence data belongs each time an element is acquired; and classification means for classifying the sequence data into one of the plurality of classes by comparing the index with a threshold, the information processing method calculating the risk of misclassifying the sequence data based on the index by calculating backwards from the final time when all of the elements are acquired, the risk at each time, and calculating the threshold so as to minimize the risk.
  • the recording medium described in Supplementary Note 8 is an information processing method for calculating the threshold used by an information processing device including: acquisition means for sequentially acquiring a finite number of elements included in sequence data; index calculation means for calculating an index indicating to which of a plurality of classes the sequence data belongs each time an element is acquired; and classification means for classifying the sequence data into one of the plurality of classes by comparing the index with a threshold, the information processing method comprising: calculating the risk of misclassifying the sequence data based on the index by calculating backward from the final time when all of the elements are acquired, the risk at each time, and calculating the threshold so as to minimize the risk, the information processing method being recorded on the recording medium.
  • the computer program described in Supplementary Note 9 is an information processing method for calculating the threshold used by an information processing device including: acquiring means for sequentially acquiring a finite number of elements included in sequence data; index calculating means for calculating an index indicating to which of a plurality of classes the sequence data belongs each time an element is acquired; and classification means for classifying the sequence data into one of the plurality of classes by comparing the index with a threshold, the computer program causing a computer to execute the information processing method, the information processing method including: calculating the risk of misclassifying the sequence data based on the index by calculating backwards from the final time when all of the elements are acquired, the risk at each time, and calculating the threshold so as to minimize the risk.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
PCT/JP2023/034766 2023-09-25 2023-09-25 情報処理装置、情報処理方法、及び記録媒体 Ceased WO2025069149A1 (ja)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2025548033A JPWO2025069149A1 (cg-RX-API-DMAC7.html) 2023-09-25 2023-09-25
PCT/JP2023/034766 WO2025069149A1 (ja) 2023-09-25 2023-09-25 情報処理装置、情報処理方法、及び記録媒体

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2023/034766 WO2025069149A1 (ja) 2023-09-25 2023-09-25 情報処理装置、情報処理方法、及び記録媒体

Publications (1)

Publication Number Publication Date
WO2025069149A1 true WO2025069149A1 (ja) 2025-04-03

Family

ID=95203019

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2023/034766 Ceased WO2025069149A1 (ja) 2023-09-25 2023-09-25 情報処理装置、情報処理方法、及び記録媒体

Country Status (2)

Country Link
JP (1) JPWO2025069149A1 (cg-RX-API-DMAC7.html)
WO (1) WO2025069149A1 (cg-RX-API-DMAC7.html)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7577709B1 (en) * 2005-02-17 2009-08-18 Aol Llc Reliability measure for a classifier
WO2021229662A1 (ja) * 2020-05-11 2021-11-18 日本電気株式会社 判定装置、判定方法および記録媒体
WO2023148846A1 (ja) * 2022-02-02 2023-08-10 日本電気株式会社 情報処理装置、情報処理方法、及び記録媒体

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7577709B1 (en) * 2005-02-17 2009-08-18 Aol Llc Reliability measure for a classifier
WO2021229662A1 (ja) * 2020-05-11 2021-11-18 日本電気株式会社 判定装置、判定方法および記録媒体
WO2023148846A1 (ja) * 2022-02-02 2023-08-10 日本電気株式会社 情報処理装置、情報処理方法、及び記録媒体

Also Published As

Publication number Publication date
JPWO2025069149A1 (cg-RX-API-DMAC7.html) 2025-04-03

Similar Documents

Publication Publication Date Title
JP7740389B2 (ja) 情報処理装置、情報処理方法、及び記録媒体
WO2023272852A1 (zh) 通过决策树模型对用户进行分类的方法、装置、设备和存储介质
CN112149615A (zh) 人脸活体检测方法、装置、介质及电子设备
CN109271929B (zh) 检测方法和装置
CN109598250B (zh) 特征提取方法、装置、电子设备和计算机可读介质
CN113722409B (zh) 空间关系的确定方法、装置、计算机设备及存储介质
CN113505848A (zh) 模型训练方法和装置
JP2021516824A (ja) 群集レベル推定のための方法、システム及びプログラム
CN113762455B (zh) 检测模型训练方法、单字检测方法、装置、设备及介质
CN115004245A (zh) 目标检测方法、装置、电子设备和计算机存储介质
CN115147474A (zh) 点云标注模型的生成方法、装置、电子设备及存储介质
CN114360053B (zh) 一种动作识别方法、终端及存储介质
CN118279624A (zh) 牙齿识别分类方法、电子设备和存储介质
CN112381709B (zh) 图像处理方法、模型训练方法、装置、设备和介质
CN112668491B (zh) 辅助阅读方法、计算机设备及存储装置
CN112967351A (zh) 图像生成方法、装置、电子设备和存储介质
CN110288668A (zh) 图像生成方法、装置、计算机设备及存储介质
CN111754604A (zh) 出行轨迹相似性确定方法及相关设备
WO2020244076A1 (zh) 人脸识别方法、装置、电子设备及存储介质
KR20210136724A (ko) 다중 스케일 예측 손실 함수를 이용한 오디오 신호 학습장치 및 방법
WO2025069245A1 (ja) 情報処理装置、情報処理方法、及び記録媒体
US12111244B2 (en) Method for calculating a density of stem cells in a cell image, electronic device, and storage medium
CN110619600A (zh) 神经网络模型训练方法及装置、存储介质及电子设备
WO2025069242A1 (ja) 情報処理装置、情報処理方法、及び記録媒体
WO2024079853A1 (ja) 情報処理装置、情報処理方法、及び記録媒体

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: 23954126

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2025548033

Country of ref document: JP

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 2025548033

Country of ref document: JP

NENP Non-entry into the national phase

Ref country code: DE