Information processing device, information processing method, and recording medium

Abstract

This information processing device comprises: an acquisition means for sequentially acquiring a plurality of elements included in series data; and a calculation means for calculating an index relating to the series data by inputting the acquired elements to a model that carries out stochastically fluctuating output with respect to the same input. The calculation means calculates the index a plurality of times by carrying out repeated input in which the same input is input to the model a plurality of times from when one element is acquired to when the next element is acquired. Such an information processing device makes it possible to appropriately calculate an index regardless of the frequency of acquiring series data.

Description

Information Processing Apparatus, Information Processing Method, and Recording Medium

[0001] This disclosure relates to the technical field of information processing apparatuses, information processing methods, and recording media.

[0002] Techniques for classifying series data are known. For example, Patent Document 1 discloses a technique for performing class classification based on a likelihood ratio calculated from a plurality of elements included in series data, and a technique for learning a model for calculating the likelihood ratio.

[0003] International Publication No. 2023 / 148846

[0004] This disclosure aims to provide an information processing apparatus, an information processing method, and a recording medium for improving the techniques disclosed in prior art documents.

[0005] One aspect of the information processing apparatus of this disclosure includes an acquisition unit that sequentially acquires a plurality of elements included in series data, and a calculation unit that calculates an index related to the series data by inputting the acquired elements into a model that outputs a probabilistically fluctuating output for the same input. The calculation unit calculates the index a plurality of times by repeatedly inputting the same input into the model a plurality of times during the period from when one element is acquired until the next element is acquired.

[0006] One aspect of the information processing method of this disclosure is that at least one computer sequentially acquires a plurality of elements included in series data, calculates an index related to the series data by inputting the acquired elements into a model that outputs a probabilistically fluctuating output for the same input, and calculates the index a plurality of times by repeatedly inputting the same input into the model a plurality of times during the period from when one element is acquired until the next element is acquired when calculating the index.

[0007] One aspect of the recording medium of this disclosure contains a computer program that causes at least one computer to execute an information processing method, which involves sequentially acquiring a plurality of elements contained in sequential data, inputting the acquired elements into a model that produces a probabilistically fluctuating output for the same input, thereby calculating an index related to the sequential data, and calculating the index multiple times by repeated input, where the same input is given to the model multiple times between the acquisition of one element and the acquisition of the next element.

[0008] This is a block diagram showing the hardware configuration of the first information processing device. This is a block diagram showing the functional configuration of the first information processing device. This is a flowchart showing the operation flow of the first information processing device. This is a schematic diagram showing the configuration of the index calculation model in the second information processing device. This is a block diagram showing the functional configuration of the third information processing device. This is a flowchart showing the operation flow of the third information processing device. This is a graph showing an example of a likelihood ratio calculated in the third information processing device. This is a graph showing an example of a likelihood ratio calculated in the fourth information processing device. This is a graph (1) showing an example of a likelihood ratio calculated in a modified version of the fourth information processing device. This is a graph (2) showing an example of a likelihood ratio calculated in a modified version of the fourth information processing device. This is a graph showing an example of a likelihood ratio calculated in the fifth information processing device. This is a graph showing an example of a likelihood ratio calculated in a modified version of the fifth information processing device. This is a graph showing an example of a likelihood ratio calculated in the sixth information processing device.

[0009] The following describes embodiments of the information processing device, information processing method, and recording medium with reference to the drawings.

[0010] <First Embodiment> The first information processing device will be described with reference to Figures 1 to 3.

[0011] (Hardware Configuration) First, the hardware configuration of the first information processing device will be described with reference to Figure 1. Figure 1 is a block diagram showing the hardware configuration of the first information processing device.

[0012] As shown in Figure 1, the first information processing device 10 includes a processor 11, a RAM (Random Access Memory) 12, a ROM (Read Only Memory) 13, a storage device 14, an input device 15, and an output device 16. The processor 11, RAM 12, ROM 13, storage device 14, input device 15, and output device 16 are all connected via a data bus 17. Note that the data bus 17 may be an interface other than a data bus (for example, LAN or USB).

[0013] The processor 11 reads a computer program. For example, the processor 11 is configured to read a computer program stored in at least one of the RAM 12, ROM 13, and storage device 14. Alternatively, 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 also obtain (i.e., read) a computer program from a device (not shown) located outside the first information processing device 10 via a network interface. The processor 11 performs various processes by executing the read computer program. When the processor 11 executes the read computer program, a functional block related to the processing performed by the first information processing device 10 is realized within the processor 11. That is, the processor 11 may function as a controller that performs various controls in the first information processing device 10.

[0014] 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 consist of one of these, or it may be configured to use multiple of them in parallel.

[0015] RAM 12 temporarily stores computer programs executed by processor 11. RAM 12 also temporarily stores data that processor 11 uses temporarily while executing computer programs. RAM 12 may be, for example, D-RAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory). Alternatively, other types of volatile memory may be used instead of RAM 12.

[0016] ROM 13 stores computer programs executed by processor 11. ROM 13 may also store other static data. ROM 13 may be, for example, P-ROM (Programmable Read Only Memory) or EPROM (Erasable Read Only Memory). Alternatively, other types of non-volatile memory may be used instead of ROM 13.

[0017] The storage device 14 stores data that the first information processing device 10 stores long-term. The storage device 14 may also operate as a temporary storage device for the processor 11. The storage device 14 may store computer programs executed by the processor 11. The storage device 14 may include, for example, at least one of a hard disk drive, a magneto-optical disk drive, an SSD (Solid State Drive), and a disk array device.

[0018] The input device 15 is a device that receives input instructions from the user of the first information processing device 10. The input device 15 may include, for example, at least one of a keyboard, mouse, touch panel, and stylus. The input device 15 may also be a device capable of voice input, for example, including a microphone.

[0019] The output device 16 is a device that outputs information related to the first information processing device 10 to the outside. For example, the output device 16 may be a display device (e.g., a display or monitor) capable of displaying information related to the first information processing device 10. Alternatively, the output device 16 may be a speaker or the like capable of outputting audio information related to the information processing device 10.

[0020] The first information processing device 10 may be configured to include some of the components described in Figure 1. For example, the first information processing device 10 may be configured to include only the processor 11, RAM 12, and ROM 13 from the components described above. In this case, the storage device 14, input device 15, and output device 16 may each be provided as external devices to the first information processing device 10. Furthermore, some of the arithmetic functions of the first information processing device 10 may be implemented by an external server or cloud.

[0021] (Functional Configuration) Next, the functional configuration of the first information processing device 10 will be described with reference to Figure 2. Figure 2 is a block diagram showing the functional configuration of the first information processing device.

[0022] In Figure 2, the first information processing device 10 is configured as a device that sequentially acquires elements included in sequential data and calculates an index related to the sequential data. The first information processing device 10 is configured to include an acquisition unit 110 and a calculation unit 120 as components for realizing its function. Note that each of the acquisition unit 110 and the calculation unit 120 may be a processing block realized by the processor 11 (see Figure 1) described above.

[0023] The acquisition unit 110 is configured to acquire sequential data. Sequential data here refers to data containing multiple elements arranged in a predetermined order, such as time-series data. More specific examples of sequential data include, but are not limited to, video data, audio data, or subdivided image data. The acquisition unit 110 is configured to acquire multiple elements included in the sequential data sequentially. For example, the acquisition unit 110 may be configured to acquire the elements included in the sequential data one by one in order. More specifically, the acquisition unit 110 may be configured to acquire frame images included in video data one frame at a time each time a picture is taken. The acquisition unit 110 may acquire data directly from any data acquisition device (e.g., a camera or microphone), or it may read data that has been acquired in advance by a data acquisition device and stored in storage. When acquiring data from a camera, the acquisition unit 110 may be configured to acquire data from each of multiple cameras. The elements of the sequential data acquired by the acquisition unit 110 are output to the calculation unit 120.

[0024] The calculation unit 120 is configured to calculate an index related to the series data. The index here is a value that indicates the characteristics of the series data, and may be, for example, a value that indicates the likelihood of the class to which the series data belongs. The calculation unit 120 calculates the index from each element of the series data acquired by the acquisition unit 110. If multiple elements have already been acquired, the calculation unit 120 may calculate the index based on the acquired multiple elements. For example, the calculation unit 120 may calculate the index based on all of the acquired multiple elements. Alternatively, the calculation unit 120 may calculate the index based on some of the acquired multiple elements. The calculation unit 120 calculates the index by inputting the elements acquired from the acquisition unit 110 into the index calculation model 200. The index calculation model 200 may be a model that has been pre-trained. In particular, the index calculation model 200 according to this embodiment produces a probabilistically fluctuating output for the same input. Therefore, even if the same elements are input into the index calculation model 200, the index output from the index calculation model 200 may be different.

[0025] The calculation unit 120 may calculate an index each time it acquires an element included in the data series acquired by the acquisition unit 110. That is, when the acquisition unit 110 acquires one element, the calculation unit 120 inputs the acquired element into the index calculation model 200 and calculates an index, and then when the acquisition unit 110 acquires another element, it inputs the acquired element into the index calculation model 200 and calculates a new index. In addition, the calculation unit 120 according to this embodiment makes the same input to the index calculation model 200 multiple times between the acquisition of one element and the acquisition of the next element (hereinafter, this operation will be appropriately referred to as "repeated input"). When the calculation unit 120 makes repeated inputs, the index will be calculated multiple times between the acquisition of one element and the acquisition of the next element. That is, more indexes will be calculated than the number of times elements are acquired. In repeated input, the same element is input to the metric calculation model 200 multiple times. However, as mentioned above, the metric calculation model 200 produces probabilistically fluctuating outputs, so the calculated metrics will each have different values. The number of repeated inputs can be set in advance, for example, according to the length of time required between the acquisition of one element and the acquisition of the next element (hereinafter referred to as the "acquisition interval" as appropriate). The number of repeated inputs can be set to increase as the acquisition interval increases.

[0026] (Operation Flow) Next, the operation flow of the first information processing device 10 will be explained with reference to Figure 3. Figure 3 is a flowchart showing the operation flow of the first information processing device.

[0027] As shown in Figure 3, when the operation of the first information processing device 10 is started, the acquisition unit 110 first acquires one element included in the sequence data (step S101). Then, the calculation unit 120 calculates an index by inputting some or all of the multiple elements acquired by the acquisition unit 110 into the index calculation model 200 (step S102).

[0028] Next, the calculation unit 120 outputs the calculated index (step S103). The index output here may be used for judgment processing related to the series data. For example, the index may be used for classifying the series data. In this case, the information processing device 10 may be configured to have a function to perform class classification based on the index. Class classification based on the index will be explained in detail in other embodiments described later.

[0029] After the calculation unit 120 outputs an index, the first information processing device 10 determines whether or not to terminate the repeated input by the calculation unit 120 (step S104). For example, the first information processing device 10 only needs to determine whether or not the repeated input has been performed up to a predetermined number of times. If the repeated input is not terminated (step S104: NO), the process from step S102, which has already been described, is executed again. That is, the same elements that were input into the index calculation model 200 immediately before are input into the index calculation model 200 again, and the process of outputting the index calculated by the index calculation model 200 is repeatedly executed.

[0030] On the other hand, if repeated input is terminated (step S104: YES), the first information processing device 10 determines whether or not to terminate the acquisition of elements (step S105). In other words, the first information processing device 10 determines whether or not to continue calculating the index. For example, the first information processing device 10 may determine to terminate the acquisition of elements when a determination result regarding the series data is obtained based on the calculated index.

[0031] If the acquisition of elements is not terminated (step S105: NO), the process from step S101, as already described, will be executed again. In this case, the acquisition unit 110 will acquire new elements and calculate new indicators. On the other hand, if the acquisition of elements is terminated (step S105: YES), the series of processes described above will be terminated.

[0032] (Technical Effects) Next, the technical effects obtained by the first information processing device 10 will be explained.

[0033] As explained in Figures 1 to 3, in the first information processing device 10, the index is calculated even during the period between acquiring elements from sequential data by repeatedly inputting to the index calculation model 200. In this embodiment in particular, because the index calculation model 200 produces a probabilistically fluctuating output, different indexes are calculated even when the same elements are repeatedly input. In this way, it is possible to calculate multiple significant indexes based on the elements acquired up to that point, even during the period before the next element is acquired. Therefore, for example, when performing judgment processing on sequential data based on the index, multiple decisions can be made even during the period before the next element is acquired. For this reason, it is possible to output the judgment result earlier compared to when the index is calculated only when an element is acquired (i.e., when repeated input is not performed).

[0034] While it may be possible to achieve a similar effect by shortening the element acquisition interval, there are limits to how much the acquisition interval can be shortened. For example, when acquiring video data as sequential data, the acquisition interval for each element (one frame of image) depends on the video's frame rate. For example, if the video was shot at 30 FPS, the acquisition interval would be approximately 30 ms, and unless repeated input is performed, decision-making can only be done at that interval. However, in this embodiment, as described above, metrics are calculated even during the period until the next element is acquired. Therefore, even when there are limitations on the element acquisition interval, more metrics can be calculated and decisions can be made.

[0035] <Second Embodiment> The second information processing device 10 will be described with reference to Figure 4. The second information processing device 10 differs from the first information processing device 10 described above in some configurations and operations, but other parts may be the same as those of the first information processing device 10. For this reason, the parts that differ from the first embodiment will be described in detail below, and explanations of other overlapping parts will be omitted as appropriate.

[0036] (Indicator Calculation Model) First, with reference to Figure 4, the indicator calculation model 200 used by the calculation unit 120 of the second information processing device 10 will be explained in detail. Figure 4 is a schematic diagram showing the configuration of the indicator calculation model in the second information processing device.

[0037] As shown in Figure 4, the index calculation model 200 in the second information processing device 10 is configured as a neural network model. The index calculation model 200 comprises an input layer, multiple hidden layers (intermediate layers), and an output layer. For the sake of explanation, here we have given an example where the index calculation model 200 has four layers, but the number of layers that the index calculation model 200 comprises is not particularly limited.

[0038] In the index calculation model 200 of the second information processing device 10, the memory length (i.e., memory retention time) of each neuron is learned for each neuron. That is, the memory length of the index calculation model 200 will be a different value for each neuron. More specifically, the index calculation model 200 may be, for example, SNNs (Spiking Neural Networks) designed based on the physiology of nerve cells. In this case, the memory length of a neuron will be a value determined based on the membrane potential time constant (hereinafter referred to as "time constant" as appropriate), and the longer the time constant, the longer the memory will remain. The index calculation model 200 calculates an index based on a number of elements determined based on the time constant. That is, the number of elements used when calculating the index will be a number corresponding to the time constant which differs for each neuron.

[0039] If the index calculation model 200 is an SNN, its behavior will be as shown in equation (1) below, for example.

[0040]

[0041] In equation (1) above, V(t) is the membrane potential, Vrest is the resting membrane potential, I(t) is the input, and τ is the input. ｍ This is the membrane potential time constant.

[0042] membrane potential time constant τ ｍ To intuitively understand its behavior, for example, we can set the input I(t) = 0 and derive the solution as shown in equation (2) below.

[0043]

[0044] Looking at the above formula (2), it can be seen that it exponentially decays from the initial state V(0) towards the resting membrane potential Vrest. Also, the membrane potential time constant τ ｍ It can be seen that the larger it is, the longer it takes to reach the resting membrane potential Vrest.

[0045] (Technical effect) Next, the technical effect obtained by the second information processing device 10 will be described.

[0046] As described in FIG. 4, in the second information processing device 10, an index calculation model 200 configured as a neural network model calculates an index related to the series data based on the number of elements determined based on the time constant learned for each neuron. In this way, it is possible to set the number of elements used when calculating the index to an appropriate number for each neuron. Therefore, it is possible to calculate an index with high accuracy while suppressing an increase in the amount of calculation when calculating the index.

[0047] For example, when all neurons have the same memory length, there is a risk that the number used to calculate the index will increase, making the processing redundant. Also, a method of decaying the memory length over time can be considered, but in that case too, the processing becomes redundant by performing the same operation on all neurons, or there is a risk that necessary information will be deleted. However, in this embodiment, as described above, the index calculation model 200 is learned to have a memory length determined based on the time constant learned for each neuron. Therefore, it is possible to suppress the above-mentioned inconveniences.

[0048] When using SNNs for the index calculation model 200, by using a DVS (Dynamic Vision Sensor), high-speed data acquisition and high-speed calculation can be guaranteed. For example, SNNs using a DVS are faster than RGB camera-based methods that process the entire image every time because they have the feature of "performing information processing only at the locations where changes occur".

[0049] <Third Embodiment> The third information processing device 10 will be described with reference to Figures 5 to 7. The third information processing device 10 differs from the first and second information processing devices 10 described above in some configurations and operations, while other parts may be the same as those of the first and second information processing devices 10. Therefore, the following will explain in detail the parts that differ from the embodiments already described, and will omit explanations of other overlapping parts as appropriate.

[0050] (Functional Configuration) First, the functional configuration of the third information processing device 10 will be explained with reference to Figure 5. Figure 5 is a block diagram showing the functional configuration of the third information processing device. Note that in Figure 5, the same reference numerals are used for elements similar to those shown in Figure 2.

[0051] In Figure 5, the third information processing device 10 is configured to include an acquisition unit 110, a calculation unit 120, and a determination unit 130 as components for realizing its function. That is, the third information processing device 10 further includes a determination unit 130 in addition to the configuration described in the first embodiment (see Figure 2). The determination unit 130 may be a processing block realized by the processor 11 (see Figure 1) described above.

[0052] The calculation unit 120 in the third information processing device 10 is configured to calculate a likelihood ratio, which indicates the likelihood of the class to which the sequence data belongs, as an index. The calculation unit 120 calculates the likelihood ratio using two or more consecutive elements from among the multiple elements acquired by the acquisition unit 110. More specifically, the calculation unit 120 calculates the likelihood ratio indicating the likelihood of the class to which the sequence data belongs by inputting two or more consecutive elements from among the multiple elements included in the sequence data into the index calculation model 200. For example, the calculation unit 120 may calculate the likelihood ratio using newly acquired elements and previously acquired elements or previously calculated likelihood ratios. In this case, the calculation unit 120 may include a storage unit for storing previously acquired elements or previously calculated likelihood ratios. The likelihood ratio may be a log-likelihood ratio (LLR). The log-likelihood ratio may be obtained, for example, by converting the posterior probabilities output when elements are input into the index calculation model 200 into prior probabilities. In this case, the temporal integrator that calculates the posterior probabilities may be constructed using the SNNs already described.

[0053] The determination unit 130 is configured to output a determination result regarding the sequence data based on the likelihood ratio calculated by the calculation unit 120. Specifically, the determination unit 130 classifies the sequence data into one of several classes based on the likelihood ratio. The determination unit 130 may classify the sequence data into one of several classes by comparing the likelihood ratio calculated by the calculation unit 120 with a pre-set threshold. For example, the determination unit 130 classifies the sequence data into class A if the likelihood ratio exceeds the threshold corresponding to class A. Also, the determination unit 130 classifies the sequence data into class B if the likelihood ratio exceeds the threshold corresponding to class B. The determination unit 130 may, for example, classify whether the face of a person in an image is a real face or a fake face (for example, an impersonation using a photograph or 3D mask). Furthermore, there may be three or more classes to which the data is classified.

[0054] (Operation Flow) Next, the operation flow of the third information processing device 10 will be explained with reference to Figure 6. Figure 6 is a flowchart showing the operation flow of the third information processing device. In the following explanation, the operation in which the determination unit 130 determines the class to which the sequence data belongs based on the likelihood ratio (i.e., classifies the data) will be used as an example.

[0055] As shown in Figure 6, when the operation of the third information processing device 10 is started, the acquisition unit 110 first acquires one element included in the sequence data (step S201). Then, the calculation unit 120 inputs two or more consecutive elements from the multiple elements acquired by the acquisition unit 110 into the index calculation model 200 to calculate the likelihood ratio that indicates the likelihood of the class to which the sequence data belongs (step S202).

[0056] Next, the determination unit 130 determines whether the likelihood ratio calculated by the calculation unit 120 exceeds a preset threshold (step S203). If the determination unit 130 determines that the likelihood ratio exceeds the threshold (step S203: YES), it determines that the series data belongs to the class corresponding to the threshold (i.e., the class corresponding to the threshold that the likelihood ratio exceeded).

[0057] On the other hand, if it is determined that the likelihood ratio does not exceed the threshold (step S203: NO), the third information processing device 10 determines whether or not to terminate the repeated input by the calculation unit 120 (step S205). If the repeated input is not terminated (step S205: NO), the process from step S202, which has already been described, is executed again. That is, the same elements that were input into the indicator calculation model 200 immediately before are input into the indicator calculation model 200 again, and the process of outputting the indicator calculated by the indicator calculation model 200 is repeatedly executed.

[0058] On the other hand, if repeated input is terminated (step S205: YES), the third information processing device 10 determines whether or not to terminate the acquisition of elements (step S206). In other words, the third information processing device 10 determines whether or not to continue calculating the index. For example, the third information processing device 10 may determine to terminate the acquisition of elements if a predetermined time has elapsed since the start of acquiring sequential data (i.e., a timeout has occurred).

[0059] If the acquisition of elements is not terminated (step S206: NO), the process from step S201, as already described, will be executed again. In this case, the acquisition unit 110 will acquire new elements and calculate new indicators. On the other hand, if the acquisition of elements is terminated (step S206: YES), the series of processes described above will be terminated.

[0060] (Example of operation) Next, a specific example of the operation of the third information processing device 10 will be explained with reference to Figure 7. Figure 7 is a graph showing an example of the likelihood ratio calculated by the third information processing device.

[0061] In the example shown in Figure 7, repeated input is performed four times between the acquisition of one element and the acquisition of the next element. That is, the same element is repeatedly input to the indicator calculation model 200 four times, resulting in the calculation of four likelihood ratios. Of the four likelihood ratios calculated through repeated input, the third likelihood ratio exceeds the threshold. In this case, the determination unit 130 can determine the class to which the series data belongs at the time the third likelihood ratio is calculated.

[0062] If repeated input is not performed, the likelihood ratio is calculated when the next element is acquired. Therefore, the timing of determining the class to which the series data belongs is delayed compared to when repeated input is performed. Also, as shown in Figure 7, even if the next element is acquired, the likelihood ratio calculated at that time may not exceed the threshold. In such cases, the timing of determining the class will be further delayed.

[0063] (Technical Effects) Next, the technical effects obtained by the third information processing device 10 will be explained.

[0064] As explained in Figures 5 to 7, the third information processing device 10 is equipped with a determination unit 130. With this configuration, determination processing (for example, class classification) on sequential data can be performed using the index calculated by the calculation unit 120. In addition, the third information processing device 10 takes two or more consecutive elements from among the multiple elements contained in the sequential data as input and calculates a likelihood ratio that indicates the likelihood of the class to which the sequential data belongs. In this way, the determination unit 130 can appropriately classify the sequential data.

[0065] <Fourth Embodiment> The fourth information processing device 10 will be described with reference to Figures 8 to 10. Note that the fourth information processing device 10 differs in some operations from the first to third information processing devices 10 described above, but other parts may be the same as the first to third information processing devices 10. For this reason, the parts that differ from the embodiments already described will be explained in detail below, and explanations of other overlapping parts will be omitted as appropriate. In the following embodiments, the index calculated by the calculation unit 120 will be described as the likelihood ratio.

[0066] (Example of operation) First, a specific example of operation by the fourth information processing device 10 will be explained with reference to Figure 8. Figure 8 is a graph showing an example of a likelihood ratio calculated by the fourth information processing device.

[0067] As shown in Figure 8, the fourth information processing device 10 sets a first predetermined range around a threshold. The first predetermined range is set in advance to determine whether the likelihood ratio is approaching the threshold to a certain extent. The fourth information processing device 10 determines whether to perform repeated input based on whether the likelihood ratio calculated immediately before is within the first predetermined range. Specifically, if the likelihood ratio is outside the first predetermined range, the fourth information processing device 10 does not perform repeated input (i.e., it does not calculate the likelihood ratio until the next element is obtained). On the other hand, if the likelihood ratio is within the first predetermined range, the fourth information processing device 10 performs repeated input.

[0068] In the example shown in Figure 8, the likelihood ratio calculated when the first element is acquired is outside the first predetermined range. Therefore, repeated input is not performed between the acquisition of the first element and the acquisition of the second element. On the other hand, the likelihood ratio calculated when the second element is acquired is within the first predetermined range. Therefore, repeated input is performed between the acquisition of the second element and the acquisition of the third element.

[0069] (Modified Versions) Next, an example of the operation of a modified version of the fourth information processing device 10 will be described with reference to Figures 9 and 10. Figure 9 is a graph (1) showing an example of the likelihood ratio calculated in the modified version of the fourth information processing device. Figure 10 is a graph (2) showing an example of the likelihood ratio calculated in the modified version of the fourth information processing device.

[0070] As shown in Figures 9 and 10, in the modified version of the fourth information processing device 10, in addition to the first predetermined range described above, the decision of whether or not to perform repeated input is made based on the slope of the likelihood ratio. Specifically, in the modified version of the fourth information processing device 10, repeated input is performed when the likelihood ratio is within the first predetermined range and the slope of the likelihood ratio is moving towards the threshold. On the other hand, in the modified version of the fourth information processing device 10, even if the likelihood ratio is within the first predetermined range, repeated input is not performed if the slope of the likelihood ratio is moving away from the threshold.

[0071] In the example shown in Figure 9, the likelihood ratio calculated when the first element is acquired is outside the first predetermined range. Therefore, no repeated input is performed between the acquisition of the first element and the acquisition of the second element. Subsequently, the likelihood ratio calculated when the second element is acquired is within the first predetermined range. Furthermore, the slope of the likelihood ratio from the acquisition of the first element to the acquisition of the second element is in the direction of approaching the threshold (i.e., the positive direction in which the likelihood ratio increases). Therefore, in this case, repeated input is performed between the acquisition of the second element and the acquisition of the third element. As a result, the likelihood ratio, which was located near the threshold, changes in the direction of approaching the threshold further and exceeds the threshold during repeated input.

[0072] On the other hand, in the example shown in Figure 10, the likelihood ratio calculated when the second element is acquired is within the first predetermined range. However, the slope of the likelihood ratio from the time the first element is acquired until the time the second element is acquired is in the direction away from the threshold (i.e., in the negative direction where the likelihood ratio decreases). Therefore, in this case, repeated input is not performed between the time the second element is acquired and the time the third element is acquired. Furthermore, the likelihood ratio, which was located near the threshold when the second element was acquired, changes further away from the threshold when the third element is acquired. For this reason, class determination is not performed. (Technical Effects) Next, the technical effects obtained by the fourth information processing device 10 will be explained.

[0073] As explained in Figure 8, the fourth information processing device 10 determines whether or not to perform repeated input based on whether the likelihood ratio is within a first predetermined range. This makes it possible to perform repeated input appropriately according to the value of the most recent likelihood ratio. For example, if the likelihood ratio is within the first predetermined range (i.e., it is somewhat close to the threshold), there is a high probability that the likelihood ratio calculated afterward will exceed the threshold. Performing repeated input in such a situation can efficiently speed up the timing of class determination. On the other hand, if the likelihood ratio is outside the first predetermined range (i.e., it is far from the threshold), there is a low probability that the likelihood ratio calculated afterward will exceed the threshold. By avoiding repeated input in such a situation, the increase in computational load due to unnecessary repeated input can be suppressed.

[0074] Furthermore, as shown in the modified examples in Figures 9 and 10, by considering the slope of the likelihood ratio, it is possible to more accurately predict subsequent changes in the likelihood ratio and determine whether repeated input is necessary. For example, if the slope of the likelihood ratio is approaching the threshold, the likelihood ratio calculated afterward is also likely to approach the threshold. Performing repeated input in such situations can efficiently speed up the timing of class determination. On the other hand, if the slope of the likelihood ratio is moving away from the threshold, the likelihood ratio calculated afterward is also likely to move away from the threshold. Avoiding repeated input in such situations can suppress the increase in computational complexity due to unnecessary repeated input.

[0075] <Fifth Embodiment> The fifth information processing device 10 will be described with reference to Figures 11 and 12. The fifth information processing device 10 differs in some operations from the first to fourth information processing devices 10 described above, but other parts may be the same as those of the first to fourth information processing devices 10. For this reason, the parts that differ from the embodiments already described will be explained in detail below, and other overlapping parts will be omitted as appropriate.

[0076] (Example of operation) First, a specific example of the operation of the fifth information processing device 10 will be explained with reference to Figure 11. Figure 11 is a graph showing an example of a likelihood ratio calculated by the fifth information processing device.

[0077] As shown in Figure 11, in the fifth information processing device 10, class determination is not performed when the likelihood ratio calculated in the repeated input exceeds the threshold only once. In the fifth information processing device 10, class determination is performed when the likelihood ratio calculated in the repeated input exceeds the threshold twice. In other words, in the fifth information processing device 10, class determination is performed on the condition that the likelihood ratio exceeds the threshold multiple times.

[0078] Here, we have given an example where the number of times class determination is performed (hereinafter referred to as the "first predetermined number") is two, but the first predetermined number may be more than two. For example, the fifth information processing device 10 may perform class determination when the likelihood ratio exceeds the threshold three times. For example, the fifth information processing device 10 may perform class determination when the likelihood ratio exceeds the threshold four times. Furthermore, the first predetermined number may be subject to the condition that it must "exceed consecutively (in other words, cumulative is not acceptable)". For example, if the first predetermined number is two, class determination may be performed when the likelihood ratio exceeds the threshold two times consecutively, while class determination may not be performed when the likelihood ratio exceeds the threshold two times with an interval in between.

[0079] Furthermore, the above-mentioned condition regarding the first predetermined number of times applies only to the likelihood ratio calculated through repeated input, and does not necessarily apply to the likelihood ratio calculated when an element is acquired. In other words, for the likelihood ratio calculated when an element is acquired, class determination may be performed only when the threshold is exceeded once.

[0080] (Modified Version) Next, an example of the operation of a modified version of the fifth information processing device 10 will be described with reference to Figure 12. Figure 12 is a graph showing an example of the likelihood ratio calculated in the modified version of the fifth information processing device.

[0081] In the modified version of the fifth information processing device 10, as a general rule, class determination is performed when the likelihood ratio calculated in repeated input as described above exceeds the first predetermined number of times. However, in the modified version of the fifth information processing device 10, a second predetermined range is further set around the threshold. The second predetermined range is set in advance to determine whether the likelihood ratio is sufficiently close to the threshold. Then, in the modified version of the fifth information processing device 10, if the likelihood ratio calculated immediately before has been within the second predetermined range for the second predetermined number of consecutive times, class determination is performed when the likelihood ratio exceeds the threshold only once.

[0082] In the example shown in Figure 12, the likelihood ratios calculated in the first, second, and third iterations of the repeated input are within the second predetermined range, while the likelihood ratio calculated in the fourth iteration exceeds the threshold. If the second predetermined number of iterations is set to "3," then the likelihood ratios calculated immediately before the fourth iteration (i.e., from the first to the third iteration) are within the second predetermined range for three consecutive iterations. Therefore, the likelihood ratio calculated in the fourth iteration will be classified at the point when it exceeds the threshold once.

[0083] (Technical Effects) Next, the technical effects obtained by the fifth information processing device 10 will be explained.

[0084] As explained in Figure 11, the fifth information processing device 10 performs a class determination when the likelihood ratio exceeds a first predetermined threshold a number of times. This reduces the possibility of making an incorrect determination compared to the case where a class determination is made when the threshold is exceeded only once. Furthermore, as shown in the modified example in Figure 12, if the likelihood ratio is located within a second predetermined range a second predetermined number of times, it is sufficient to exceed the threshold only once, allowing for earlier class determination depending on the situation. For example, if the likelihood ratio is fluctuating very close to the threshold, it is considered highly likely that the likelihood ratio calculated afterward will exceed the threshold. Therefore, even if a class determination is made when the threshold is exceeded only once, the possibility of an incorrect determination result is low. Thus, early class determination can be performed while suppressing misdetermination.

[0085] <Sixth Embodiment> The sixth information processing device 10 will be described with reference to Figure 13. Note that the sixth information processing device 10 differs in some operations from the first to fifth information processing devices 10 described above, but other parts may be the same as the first to fifth information processing devices 10. For this reason, the parts that differ from each embodiment already described will be explained in detail below, and other overlapping parts will be omitted as appropriate.

[0086] (Example of operation) First, a specific example of the operation of the sixth information processing device 10 will be explained with reference to Figure 13. Figure 13 is a graph showing an example of a likelihood ratio calculated by the sixth information processing device.

[0087] The sixth information processing device 10 does not perform class determination if the likelihood ratio is oscillating, even if the likelihood ratio exceeds a threshold. Specifically, the sixth information processing device 10 determines whether the sign of the slope of the likelihood ratio calculated by repeated operation is changing beyond a predetermined frequency. If the sign of the slope of the likelihood ratio is changing beyond a predetermined frequency, it does not perform class determination even if the likelihood ratio exceeds a threshold. On the other hand, if the sign of the slope of the likelihood ratio is not changing beyond a predetermined frequency, it performs class determination at the time the likelihood ratio exceeds the threshold.

[0088] In the example shown in Figure 13, although some of the likelihood ratios calculated in the repeated input exceed the threshold, the likelihood ratios are oscillating, with states where they exceed the threshold and states where they do not exceed the threshold alternating. For this reason, the sixth information processing device 10 does not perform class determination in the repeated input.

[0089] Furthermore, the likelihood ratio calculated at the time an element is acquired may be used to perform class determination even if the previous likelihood ratio is oscillating. In other words, the likelihood ratio oscillation can be judged based only on the likelihood ratio calculated through repeated input.

[0090] (Technical Effects) Next, the technical effects obtained by the sixth information processing device 10 will be explained.

[0091] As explained in Figure 13, in the sixth information processing device 10, class determination is not performed if the likelihood ratio calculated in repeated input is fluctuating. The likelihood ratio calculated in repeated input is considered to have lower reliability compared to the likelihood ratio calculated when an element is acquired (i.e., the likelihood ratio calculated considering the newly acquired element). Therefore, by not performing class determination when the likelihood ratio is fluctuating, it is possible to reduce the possibility of making an incorrect determination.

[0092] The processing method of recording a program that operates the configuration of each embodiment in order to realize the functions of each embodiment described above on a recording medium, reading the program recorded on the recording medium as code, and executing it on a computer is also included in the scope of each embodiment. In other words, a computer-readable recording medium is also included in the scope of each embodiment. Furthermore, not only the recording medium on which the above-mentioned program is recorded, but also the program itself is included in each embodiment.

[0093] Examples of recording media that can be used include floppy disks (registered trademark), hard disks, optical disks, magneto-optical disks, CD-ROMs, magnetic tapes, non-volatile memory cards, and ROMs. Furthermore, the scope of each embodiment is not limited to programs that perform processing on the recording media alone, but also includes programs that operate on the OS and perform processing in cooperation with other software and the functions of expansion boards. In addition, the program itself may be stored on a server, and part or all of the program may be made available for download from the server to the user terminal. The program may be provided to the user in, for example, SaaS (Software as a Service) format.

[0094] <Note> The embodiments described above may also be described in the following way, but are not limited to the following.

[0095] (Note 1) The information processing device described in Note 1 comprises an acquisition means for sequentially acquiring multiple elements included in sequential data, and a calculation means for calculating an index relating to the sequential data by inputting the acquired elements into a model that produces a probabilistically fluctuating output for the same input, wherein the calculation means calculates the index multiple times by repeated input, which involves making the same input to the model multiple times between the acquisition of one element and the acquisition of the next element.

[0096] (Note 2) The information processing device described in Note 2 is the information processing device described in Note 1, wherein the model is a neural network model that is learned using the plurality of elements acquired sequentially, and the calculation means calculates the index based on a number of elements determined based on the time constant learned for each neuron of the model.

[0097] (Note 3) The information processing device described in Note 3 is the information processing device described in Note 1 or 2, wherein the index is a value indicating the likelihood of the class to which the sequence data belongs, and the calculation means calculates the index by taking two or more consecutive elements from the acquired elements as input.

[0098] (Appendix 4) The information processing device described in Appendix 4 is the information processing device described in any one of the appendices 1 to 3, further comprising determination means for outputting a determination result regarding the sequence data by comparing the index with a predetermined threshold.

[0099] (Note 5) The information processing device described in Note 5 is the information processing device described in Note 4, wherein the calculation means performs the repeated input when the index is located within a first predetermined range from the predetermined threshold.

[0100] (Note 6) The information processing device described in Note 6 is the information processing device described in Note 5, wherein the calculation means performs the repeated input when the index is located within the first predetermined range from the predetermined threshold and the slope of the index is in the direction of approaching the predetermined threshold.

[0101] (Note 7) The information processing device described in Note 7 is an information processing device described in any one of Notes 4 to 6, wherein the determination means outputs the determination result when the index calculated by the repeated operation exceeds the predetermined threshold a first predetermined number of times.

[0102] (Note 8) The information processing device described in Note 8 is the information processing device described in Note 7, wherein the determination means outputs the determination result when the index is located within a second predetermined range from the predetermined threshold for a second predetermined number of consecutive times, and the index calculated by the repeated operation exceeds the predetermined threshold only once.

[0103] (Note 9) The information processing device described in Note 9 is the information processing device described in any one of claims 4 to 8, wherein the determination means does not output the determination result if the sign of the slope of the index calculated by the repeated operation changes beyond a predetermined frequency.

[0104] (Note 10) The information processing method described in Note 10 is an information processing method in which at least one computer sequentially acquires a plurality of elements included in sequential data, inputs the acquired elements into a model that produces a probabilistically fluctuating output for the same input, calculates an index relating to the sequential data, and calculates the index multiple times by repeated input, where the same input is given to the model multiple times between the acquisition of one element and the acquisition of the next element.

[0105] (Note 11) The recording medium described in Note 13 is a recording medium on which a computer program is recorded that causes at least one computer to execute an information processing method which involves sequentially acquiring multiple elements included in sequential data, inputting the acquired elements into a model that produces a probabilistically fluctuating output for the same input to calculate an index relating to the sequential data, and calculating the index multiple times by repeated input, which involves making the same input to the model multiple times between the acquisition of one element and the acquisition of the next element.

[0106] (Note 12) The computer program described in Note 16 is a computer program that causes at least one computer to execute an information processing method which involves sequentially acquiring multiple elements contained in sequential data, inputting the acquired elements into a model that produces a probabilistically fluctuating output for the same input, thereby calculating an index related to the sequential data, and calculating the index multiple times by repeatedly inputting the same input to the model multiple times between the acquisition of one element and the acquisition of the next element.

[0107] This disclosure may be modified as appropriate, insofar as it does not contradict the gist or idea of ​​the invention as can be inferred from the claims and the specification as a whole, and information processing devices, information processing methods, and recording media with such modifications are also included in the technical idea of ​​this disclosure.

[0108] 10 Information processing device 11 Processor 12 RAM 13 ROM 14 Storage device 15 Input device 16 Output device 17 Data bus 110 Acquisition unit 120 Calculation unit 130 Judgment unit 200 Index calculation model

Claims

1. An information processing device comprising: an acquisition means for sequentially acquiring multiple elements contained in sequential data; and a calculation means for calculating an index relating to the sequential data by inputting the acquired elements into a model that produces a probabilistically fluctuating output for the same input, wherein the calculation means calculates the index multiple times by repeated input, where the same input is given to the model multiple times between the acquisition of one element and the acquisition of the next element.

2. The information processing apparatus according to claim 1, wherein the model is a neural network model learned using the plurality of elements acquired sequentially, and the calculation means calculates the index based on a number of elements determined based on a time constant learned for each neuron of the model.

3. The information processing apparatus according to claim 1 or 2, wherein the index is a value indicating the likelihood of the class to which the sequence data belongs, and the calculation means calculates the index using two or more consecutive elements from the acquired elements as input.

4. The information processing apparatus according to claim 1 or 2, further comprising determination means for outputting a determination result regarding the series data by comparing the index with a predetermined threshold.

5. The information processing apparatus according to claim 4, wherein the calculation means performs the repeated input when the index is located within a first predetermined range from the predetermined threshold.

6. The information processing apparatus according to claim 5, wherein the calculation means performs the repeated input when the index is located within the first predetermined range from the predetermined threshold and the slope of the index is in a direction approaching the predetermined threshold.

7. The information processing apparatus according to claim 4, wherein the determination means outputs the determination result when the index calculated by the repeated operation exceeds the predetermined threshold a first predetermined number of times.

8. The information processing apparatus according to claim 7, wherein the determination means outputs the determination result even if the indicator calculated by the repeated operation exceeds the predetermined threshold only once, when the indicator is located within a second predetermined range from the predetermined threshold for a second predetermined number of consecutive times.

9. The information processing apparatus according to claim 4, wherein the determination means does not output the determination result if the sign of the slope of the index calculated by the repeated operation changes beyond a predetermined frequency.

10. An information processing method comprising: at least one computer sequentially acquiring multiple elements contained in sequential data; inputting the acquired elements into a model that produces a probabilistically fluctuating output for the same input to calculate an index relating to the sequential data; and calculating the index multiple times by repeated input, where the same input is given to the model multiple times between the acquisition of one element and the acquisition of the next element.

11. A recording medium on which a computer program is stored that causes at least one computer to execute an information processing method comprising: sequentially acquiring multiple elements contained in sequential data; inputting the acquired elements into a model that produces a probabilistically fluctuating output for the same input to calculate an index relating to the sequential data; and calculating the index multiple times by repeated input, where the same input is given to the model multiple times between the acquisition of one element and the acquisition of the next element.

Images