Estimation apparatus

Abstract

To reduce annotation workload to efficiently construct a learning model capable of achieving human behavior estimation.SOLUTION: An estimation apparatus 1 includes: a data receiving unit 11 which receives, in time series, signal state data related to amplitude or phase of receiving signals from a Wi-Fi receiver; a second estimation unit 13 which inputs input data based on the time-series signal state data to a learning model to estimate human behavior on the basis of an output from the learning model; a data generation unit 14 which performs first data augmentation on the time-series signal state data to generate first training data, and performs second data augmentation on the time-series signal state data to generate second training data; and a training unit 15 which adds a pseudo-label to the first training data, on the basis of a result of inputting the first training data to the learning model, and trains the learning model using the second training data and the pseudo-label.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to an estimation device. [Background technology]

[0002] Patent Document 1 discloses a bathroom monitoring device that detects the state of a bather. This device calculates the signal strength in each of multiple areas based on the reflected waves received by a radio wave receiving unit, stores a classification model through machine learning using multiple reference data obtained based on the signal strength, and uses the classification model to determine which of multiple classes the analytical data to be analyzed belongs to. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-86935 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the device described in Patent Document 1, in order to realize a highly accurate model, annotation work on training data is required, and generation of a learning model tends to be time-consuming.

[0005] Therefore, an object of the present disclosure is to provide an estimation device that can reduce the amount of annotation work and efficiently build a learning model that can estimate human behavior. [Means for solving the problem]

[0006] The estimation device of the present disclosure includes a data receiving unit that receives signal state data related to the amplitude or phase of a received signal in a time series from a short-range wireless communication device; an estimation unit that inputs input data based on the time series signal state data into a learning model and thereby estimates human behavior based on the output of the learning model; a data generating unit that performs first data extension on the time series signal state data received by the data receiving unit to generate first training data and performs second data extension on the time series signal state data to generate second training data; and a training unit that assigns pseudo labels to the first training data based on the result of inputting the first training data into the learning model, and trains the learning model using the second training data and the pseudo labels. [Effects of the Invention]

[0007] According to the present disclosure, it is possible to reduce the amount of annotation work and efficiently build a learning model capable of estimating human behavior. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram showing the configuration of an estimation device 1 according to the present disclosure. [Figure 2] FIG. 2 is a diagram showing an example of the structure of CSI data received from each of the Wi-Fi receivers 3a, 3b, and 3c. [Figure 3] FIG. 3 is a diagram showing an example of time-series amplitude value changes over time in signal state data of one channel. [Figure 4] FIG. 4 is a diagram showing the flow of processing by the data generating section 14 and the training section 15 shown in FIG. [Figure 5] 5 shows an example of the configuration of the narrow-range learning model M1 stored in the second learning model storage unit 17 of FIG. [Figure 6] FIG. 6 is a flowchart showing the procedure of the data collection phase performed by the estimation device 1. [Figure 7]FIG. 7 is a flowchart showing the procedure of the training phase performed by the estimation device 1. [Figure 8] FIG. 8 is a flowchart showing the procedure of the estimation phase performed by the estimation device 1. [Figure 9] FIG. 9 is a diagram illustrating an example of a hardware configuration of an estimation device 1 according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0009] The present disclosure will be described with reference to the accompanying drawings. Whenever possible, the same parts are designated by the same reference numerals and redundant description will be omitted.

[0010] FIG. 1 is a diagram illustrating the configuration of an estimation device 1 according to the present disclosure. The estimation device 1 is configured to be capable of data communication with multiple Wi-Fi receivers (short-range wireless communication devices) 3a, 3b, and 3c via a communication network (not shown). For convenience, three Wi-Fi receivers are illustrated in FIG. 1, but the estimation device 1 may be configured to be capable of data communication with any number of Wi-Fi receivers. The Wi-Fi receivers 3a, 3b, and 3c are configured to be capable of receiving wireless signals from one or more Wi-Fi transmitters (parent devices) 5 via short-range wireless communication conforming to the IEEE802.11 communication standard. In this embodiment, the Wi-Fi receivers 3a, 3b, and 3c are located apart from one another indoors, for example, in different rooms indoors.

[0011] As will be described later, the estimation device 1 is a device that estimates human behavior using a learning model. That is, the estimation device 1 estimates human behavior using state information related to wireless signals acquired by Wi-Fi receivers 3a, 3b, and 3c. In this embodiment, the estimation device 1 estimates whether a person is present in any room indoors, and further estimates the behavior of the person.

[0012] The estimation device 1 is configured to include, as functional components, a data receiving unit 11, a first estimation unit 12, a second estimation unit 13, a data generation unit 14, a training unit 15, a first learning model storage unit 16, and a second learning model storage unit 17.

[0013] The data receiving unit 11 receives CSI (Channel State Information) data from each of the Wi-Fi receivers 3a, 3b, and 3c as needed (e.g., periodically). The CSI data is data related to received signals used in the physical layer of Wi-Fi communication (near-field wireless communication), generated by each of the Wi-Fi receivers 3a, 3b, and 3c, and indicates the state of the radio wave path between the transmitter and receiver. FIG. 2 shows an example of the configuration of CSI data received from each of the Wi-Fi receivers 3a, 3b, and 3c. As described above, the CSI data includes, for example, phase values ​​and amplitude values ​​of 60 channels for each timestamp indicating the time at which the state of the radio wave path was detected. The phase value of each channel (e.g., “20.19842”) indicates the phase value of the radio signal on the radio wave path of each channel, and the amplitude value of each channel (e.g., “15.68923”) indicates the amplitude value of the radio signal on the radio wave path of each channel. The CSI data includes phase and amplitude values ​​for the number of channels (e.g., 2 x 3 x 10 = 60 channels) calculated by multiplying the number of transmitting antennas (e.g., 2) by the number of receiving antennas (e.g., 3) and the number of carriers (e.g., 10).

[0014] The data receiving unit 11 references the CSI data received from each of the Wi-Fi receivers 3a, 3b, and 3c, converts the phase and amplitude values ​​of each channel into complex numbers, and generates signal state data (input data) in a format (e.g., CSV format) in which the complex numbers of each channel are arranged in time series. Figure 3 shows an example of time-series changes in amplitude values ​​in the signal state data of one channel. The data receiving unit 11 then passes the signal state data of each channel for each of the Wi-Fi receivers 3a, 3b, and 3c to the first estimating unit 12 or the data generating unit 14.

[0015] In the estimation phase, the first estimation unit 12 inputs signal state data of multiple channels for each Wi-Fi receiver 3a, 3b, and 3c into the global learning model, which is a one-dimensional convolutional neural network (CNN) model, using the global learning model for each Wi-Fi receiver 3a, 3b, and 3c stored in the first learning model storage unit 16. Based on the output of the global learning model, the first estimation unit 12 determines whether a person is present in each room in which the Wi-Fi receiver 3a, 3b, and 3c is located. In the training phase, the first estimation unit 12 trains the global learning model by supervised learning using the signal state data of multiple channels for each Wi-Fi receiver 3a, 3b, and 3c for each global learning model stored in the first learning model storage unit 16. Then, the first estimation unit 12 stores parameters such as weighting coefficients of the global learning model updated by the training in the first learning model storage unit 16.

[0016] The second estimation unit 13 targets the Wi-Fi receivers 3a, 3b, and 3c for which the first estimation unit has estimated that a person is present in the room. The second estimation unit 13 inputs signal state data of multiple channels of the corresponding Wi-Fi receivers 3a, 3b, and 3c into a narrow-range learning model, which is a one-dimensional convolutional neural network (CNN) model stored in the second learning model storage unit 17, and determines the type of human behavior in the room where the corresponding Wi-Fi receivers 3a, 3b, and 3c are located based on the output of the narrow-range learning model. The second estimation unit 13 then outputs the estimated type of human behavior. The output may be performed by transmitting the data to an external device such as a terminal, or may be performed to an output device or storage device of the estimation device 1.

[0017] In the training phase for training the narrow-band learning model, the data generation unit 14 collects signal state data for each channel of the multiple Wi-Fi receivers 3a, 3b, and 3c, and generates training data for training the narrow-band learning model based on the collected signal state data. That is, the data generation unit 14 extracts training data for supervised learning and training data for unsupervised learning at a predetermined ratio from the multiple signal state data. Labels (correct values) indicating types of human behavior are added to this training data for supervised learning based on input from the operator of the estimation device 1 (annotation is performed to label the correct values).

[0018] The data generation unit 14 then generates first training data and second training data using the extracted multiple sequences of training data for unsupervised learning as follows. That is, the data generation unit 14 generates multiple sequences of first training data by performing a weak data transformation process (first data augmentation) on the multiple sequences of training data for unsupervised learning. As the first data augmentation, a pattern mixing augmentation method that combines multiple time series data patterns to generate a new pattern (see the paper "Time series data augmentation for neural networks by time warping with a discriminative teacher" in: ICPR; (2021)) may be used. That is, the data generation unit 14 selects one sequence of data from the multiple sequences of training data for unsupervised learning, extracts the most discriminative sequence of data from the other sequences of training data for unsupervised learning, and generates pairs of data from the one sequence and data from the other sequences. At this time, to determine discriminability, the DTW (Dynamic Time Warping) distance (see Internet URL: https: / / arxiv.org / abs / 1606.01601) between the two series of data is calculated, and the two series of data with the largest distance are paired. Furthermore, the data generation unit 14 transforms one series of data of the pair so that it approaches the pattern (characteristics) of the other series of data, thereby transforming it into data having the characteristics of the two series. The data generation unit 14 generates multiple series of first training data by repeating the above process for multiple series of training data for unsupervised learning.

[0019] Meanwhile, the data generation unit 14 generates multiple sequences of second training data by performing a strong data transformation process (second data augmentation) on multiple sequences of training data for unsupervised learning. For the second data augmentation, magnitude warping, which can add variation while preserving the characteristics of time-series data (see the paper "Data augmentation of wearable sensor data for Parkinson's disease monitoring using convolutional neural networks", in: ACM ICMI; pp. 216-220 (2017)), may be used. This method is a time-series data augmentation method that changes the signal magnitude using cubic spline interpolation to create a smooth curve. This augmentation method generates data that has undergone strong data transformation, changing the signal magnitude while retaining some of the characteristics of the time-series data.

[0020] In the training phase for training the narrow-range learning model, the training unit 15 trains the narrow-range learning model using three types of training data: the training data for supervised learning (third training data) generated by the data generation unit 14, the first training data of multiple sequences, and the second training data of multiple sequences.

[0021] First, the training unit 15 performs a first data augmentation process, which is a weak data transformation process, on the labeled training data for supervised learning of multiple sequences, and then inputs the data to the narrow-range learning model stored in the second learning model storage unit 17, and outputs the output value q i m (m indicates the data sequence, and i indicates the class, which is the type of human behavior.) Here, the output value q of the narrow learning model is obtained. i m is an output vector of the Softmax function, and each element of the output vector represents the probability value of each class. Then, the training unit 15 calculates the correct value y i mand the obtained output value q i m Based on this, the following formula (1):

number

[0022] The training unit 15 inputs the first training data of each sequence into the narrow-range learning model stored in the second learning model storage unit 17, and calculates the output value q i m (m indicates a data sequence, and i indicates a class that is a type of human behavior). Next, the training unit 15 obtains the obtained output value q i m The maximum value is extracted from the probability values ​​of each class included in p. If the maximum value exceeds a threshold, the maximum value is replaced with 1, and the other probability values ​​are set to 0. i m The training unit 15 assigns the pseudo label p i m The assignment of the first training data is repeated for multiple sequences of the first training data.

[0023] Furthermore, the training unit 15 inputs each of the multiple series of second training data to the narrow-range learning model stored in the second learning model storage unit 17, and calculates the output value q of the narrow-range learning model. i m (m indicates a data sequence, and i indicates a class that is a type of human behavior). Then, the training unit 15 obtains each of the obtained output values ​​q i m and the respective output values ​​q i mThe pseudo-label p assigned to the first training data of the data sequence corresponding to the second training data on which i m Based on this, the following equation (2) is used:

number

[0024] Furthermore, the training unit 15 calculates an added loss value by weighting and adding the loss value L2 calculated based on the training data for supervised learning and the loss value L1 calculated based on the first training data of multiple sequences and the second training data of multiple sequences according to the number of epochs. Specifically, the training unit 15 calculates an added loss value by using the following formula (3);

number

[0025] 4 shows the flow of processing by the data generation unit 14 and the training unit 15. The data generation unit 14 generates first training data TD1 that has been subjected to weak data transformation processing and second training data TD2 that has been subjected to strong data transformation processing based on the training data TDA for unsupervised learning. The training unit 15 inputs the first training data TD1 of multiple sequences into the narrow-band learning model M1, thereby generating an output value q i m is obtained, and its output value q i m Based on the above, the first training data TD1 of each of the multiple sequences is assigned a pseudo label p i m In addition, the training unit 15 inputs a plurality of sequences of second training data TD2 to the narrow-band learning model M1, thereby generating an output value q i m are obtained, and the pseudo labels p i m and the output value q based on the second training data TD2 of each sequence. i m The loss value L1 is calculated using

[0026] On the other hand, the training unit 15 augments the training data TDB for supervised learning of multiple sequences and then inputs it to the narrow-band learning model M1, thereby obtaining an output value q i m is obtained, and its output value q i m and the correct answer y based on the labels attached to the training data TDB for supervised learning. i m The loss value L2 is calculated using the weighted sum of the two loss values ​​L1 and L2 by the training unit 15. The training unit 15 then calculates an added loss value L by weighting the two loss values ​​L1 and L2 with weighting according to the number of epochs, and the narrow-range learning model is trained using this added loss value L.

[0027] 5 shows an example of the configuration of the narrow-range learning model M1 stored in the second learning model storage unit 17. The narrow-range learning model M1 has a configuration in which the following are connected in this order: a convolutional layer ML1 that performs convolutional operations on time-series input data, a maximum value pooling layer ML2 that aggregates to the maximum value, a convolutional layer ML3, a maximum value pooling layer ML4, a convolutional layer ML5, a global average pooling layer ML6 that aggregates to the average value, a dense layer ML7 which is a so-called fully connected layer, a dropout layer ML8 that randomly deactivates nodes, and a softmax layer ML9 that generates output values. The global learning model stored in the first learning model storage unit 16 may also have a similar configuration.

[0028] The following describes the processing procedures performed by the estimation device 1 configured as described above. Fig. 6 is a flowchart showing the procedures of the data collection phase performed by the estimation device 1, Fig. 7 is a flowchart showing the procedures of the training phase performed by the estimation device 1, and Fig. 8 is a flowchart showing the procedures of the estimation phase performed by the estimation device 1.

[0029] As shown in FIG. 6, in the data collection phase, first, the Wi-Fi transmitter 5 periodically transmits a ping signal to each of the Wi-Fi receivers 3a, 3b, and 3c, triggering the acquisition of CSI data (step S01). In response, each of the Wi-Fi receivers 3a, 3b, and 3c acquires CSI data upon receiving the ping signal (step S02). Then, each of the Wi-Fi receivers 3a, 3b, and 3c transmits the acquired CSI data to the estimation device 1 (step S03). Then, the estimation device 1 processes the received CSI data into multiple sequences of signal state data in CSV format (step S04). Finally, the estimation device 1 performs deletion of null data and interpolation on the multiple sequences of signal state data (step S05).

[0030] As shown in FIG. 7, in the training phase, the estimation device 1 labels (annotates) some of the multiple sequences of signal state data (step S101). Next, the estimation device 1 divides the multiple sequences of signal state data into training data, validation data, and test data (step S102). Then, training (global learning) is performed on the global learning model for each Wi-Fi receiver 3a, 3b, and 3c using the training data received from each Wi-Fi receiver 3a, 3b, and 3c (step S103). Finally, training (local learning) is performed on the narrow-area learning model using the training data for each room received from the Wi-Fi receivers 3a, 3b, and 3c, with data augmentation (step S104).

[0031] As shown in FIG. 8, in the estimation phase, the estimation device 1 estimates a room in which a person is present using a global learning model for each of the Wi-Fi receivers 3a, 3b, and 3c based on the signal condition data received from each of the Wi-Fi receivers 3a, 3b, and 3c (step S201). Next, the estimation device 1 selects a Wi-Fi receiver corresponding to the estimated room from among the Wi-Fi receivers 3a, 3b, and 3c (step S202). Thereafter, the estimation device 1 estimates the type of person's behavior using a narrow-area learning model based on the signal condition data received from the selected Wi-Fi receiver (step S203). Finally, the estimation device 1 outputs a final estimation result (step S204).

[0032] Next, we will explain the effects of the estimation device 1 of the present disclosure. The estimation device 1 of the present disclosure includes a data receiving unit 11 that receives signal state data related to the amplitude or phase of received signals in a time series from Wi-Fi receivers 3 a, 3 b, and 3 c, a second estimation unit 13 that inputs input data based on the time series signal state data into a narrow-band learning model and thereby estimates human behavior based on the output of the narrow-band learning model, a data generating unit 14 that performs first data extension on the time-series signal state data received by the data receiving unit 11 to generate first training data and performs second data extension on the time-series signal state data to generate second training data, and a training unit 15 that assigns pseudo labels to the first training data based on the result of inputting the first training data to the narrow-band learning model and trains the narrow-band learning model using the second training data and the pseudo labels.

[0033] With this configuration, input data based on time-series signal state data received from Wi-Fi receivers 3a, 3b, and 3c is input to the narrow-band learning model, and human behavior is estimated based on the output of the narrow-band learning model. Here, pseudo labels obtained from first training data, in which the time-series signal state data has been subjected to first data augmentation, and second training data, in which the time-series signal state data has been subjected to second data augmentation, are used to train the narrow-band learning model. As a result, annotation work for the training data is reduced, and a narrow-band learning model capable of estimating human behavior can be efficiently constructed.

[0034] Here, in the estimation device 1 of the present disclosure, the first data extension is a process of weakly transforming the time-series signal state data, and the second data extension is a process of strongly transforming the time-series signal state data.

[0035] This allows us to build a learning model that can estimate human behavior with high accuracy even for data-transformed signal state data, thereby improving the accuracy of human behavior estimation.

[0036] In the estimation device 1 of the present disclosure, the training unit 15 calculates a loss value L1 based on the result of inputting the second training data into the narrow range learning model and the pseudo label, and updates the parameters of the narrow range learning model using the loss value L1. This makes it possible to construct a narrow range learning model so that the output of the narrow range learning model based on the second training data approaches the behavior indicated by the pseudo label, thereby achieving highly accurate estimation of human behavior.

[0037] Furthermore, in the estimation device 1 of the present disclosure, the training unit 15 further acquires third training data in which time-series signal state data is labeled with correct values, further calculates a loss value L2 based on the result of inputting the third training data into the narrow-range learning model and the correct values, and further uses the loss value L2 to update the parameters of the narrow-range learning model. This makes it possible to construct a narrow-range learning model using supervised learning based on the third training data as well, thereby achieving even more accurate estimation of human behavior.

[0038] Furthermore, in the estimation device 1 of the present disclosure, the training unit 15 calculates an added loss value L by performing weighted addition of the loss value L1 and the loss value L2, with the weighting changed according to the number of epochs, and updates the parameters of the narrow-range learning model based on the added loss value L. In this way, it is possible to construct a narrow-range learning model by appropriately combining learning using the first training data and the second training data with learning using the third training data. As a result, it is possible to realize more accurate estimation of human behavior.

[0039] Furthermore, in the estimation device 1 of the present disclosure, the first data augmentation is a process of extracting two pairs of time-series signal state data from a plurality of time-series signal state data, and transforming one of the two time-series signal state data so that it also has the characteristics of the other of the two time-series signal state data. In this case, the first data augmentation can be easily realized as a process of weakly transforming data to suit the actual environment.

[0040] Furthermore, in the estimation device 1 of the present disclosure, the second data extension is spline interpolation processing. In this case, the second data extension can be easily realized as processing that strongly transforms data.

[0041] The estimation device of the present disclosure has the following configuration.

[0042] [1] a data receiving unit that receives signal state data relating to the amplitude or phase of a received signal from the short-range wireless communication device in time series; an estimation unit that inputs input data based on the time-series signal state data into a learning model and estimates human behavior based on an output of the learning model; a data generating unit that performs first data extension on the time-series signal state data received by the data receiving unit to generate first training data, and performs second data extension on the time-series signal state data to generate second training data; a training unit that assigns pseudo labels to the first training data based on a result of inputting the first training data into the learning model, and trains the learning model using the second training data and the pseudo labels; An estimation device comprising:

[0043] [2] The first data extension is a process of weakly transforming the time-series signal condition data, The second data extension is a process of strongly transforming the time-series signal condition data. The estimation device according to [1] above.

[0044] [3] the training unit calculates a first loss function based on a result of inputting the second training data into the learning model and the pseudo label, and updates parameters of the learning model using the first loss function. The estimation device according to [1] or [2] above.

[0045] [4] the training unit further acquires third training data in which correct values ​​are labeled with respect to the time-series signal state data, further calculates a second loss function based on a result of inputting the third training data into the learning model and the correct values, and further uses the second loss function to update parameters of the learning model. The estimation device according to [3] above.

[0046] [5] the training unit calculates an additive loss function by adding the value of the first loss function and the value of the second loss function, and updates parameters of the learning model based on the additive loss function. The estimation device according to [4] above.

[0047] [6] the training unit calculates an additive loss function by performing a weighted addition of the value of the first loss function and the value of the second loss function, with the weighting changed according to the number of epochs, and updates parameters of the learning model based on the additive loss function; The estimation device according to [4] above.

[0048] [7] The first data extension is a process of extracting two pairs of time series of the signal state data from a plurality of time series of the signal state data, and transforming one of the two time series of the signal state data so that it also has characteristics of the other of the two time series of the signal state data. The estimation device according to any one of [1] to [6] above.

[0049] [8] The second data expansion is a spline interpolation process. The estimation device according to any one of [1] to [7] above.

[0050] The block diagrams used to explain the above embodiments show functional blocks. These functional blocks (components) are realized by any combination of hardware and / or software. Furthermore, the method for realizing each functional block is not particularly limited. That is, each functional block may be realized using a single device that is physically or logically coupled, or may be realized using two or more physically or logically separated devices that are connected directly or indirectly (for example, by wire, wirelessly, etc.) and these multiple devices. The functional block may also be realized by combining the single device or multiple devices with software.

[0051] Functions include, but are not limited to, judgment, determination, judgment, calculation, computation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, resolution, selection, election, establishment, comparison, assumption, expectation, consideration, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocation, mapping, and assignment. For example, a functional block (component) that performs transmission is called a transmitting unit or transmitter. As mentioned above, there are no particular limitations on how these functions are implemented.

[0052] For example, the estimation device 1 according to an embodiment of the present disclosure may function as a computer that performs the processing of the present disclosure. FIG. 9 is a diagram illustrating an example of a hardware configuration of the estimation device 1 according to an embodiment of the present disclosure. The above-described estimation device 1 may be physically configured as a computer device including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, and the like. Note that the estimation device 1 may be configured as a computer device including at least one processor such as a CPU or a GPU, or may be configured as a computer device including multiple processors, or may be configured to include multiple computer devices.

[0053] In the following description, the term "apparatus" can be interpreted as a circuit, a device, a unit, etc. The hardware configuration of the estimation apparatus 1 may be configured to include one or more of the apparatuses shown in the drawings, or may be configured to exclude some of the apparatuses.

[0054] Each function in the estimation device 1 is realized by loading predetermined software (programs) onto hardware such as the processor 1001 and memory 1002, causing the processor 1001 to perform calculations, control communication via the communication device 1004, and control at least one of reading and writing data in the memory 1002 and storage 1003.

[0055] The processor 1001 controls the entire computer by running, for example, an operating system. The processor 1001 may be configured by a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic unit, a register, etc. For example, the above-mentioned data receiving unit 11, first estimating unit 12, second estimating unit 13, data generating unit 14, training unit 15, etc. may be realized by the processor 1001.

[0056] The processor 1001 also loads programs (program codes), software modules, data, etc. from at least one of the storage 1003 and the communication device 1004 into the memory 1002 and executes various processes in accordance with the programs. The programs used are those that cause a computer to execute at least some of the operations described in the above-described embodiments. For example, the data receiving unit 11, the first estimating unit 12, the second estimating unit 13, the data generating unit 14, and the training unit 15 may be implemented by a control program stored in the memory 1002 and running on the processor 1001, and similar implementations may be used for other functional blocks. While the above-described various processes have been described as being executed by one processor 1001, they may also be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The programs may be transmitted from a network via a telecommunications line.

[0057] The memory 1002 is a computer-readable recording medium and may be configured by, for example, at least one of a read-only memory (ROM), an erasable programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a random access memory (RAM), etc. The memory 1002 may also be called a register, a cache, a main memory (primary storage device), etc. The memory 1002 can store executable programs (program codes), software modules, etc. for implementing a processing method according to one embodiment of the present disclosure.

[0058] Storage 1003 is a computer-readable recording medium, and may be, for example, at least one of an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray disc), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy disk, a magnetic strip, etc. Storage 1003 may also be referred to as an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, a server, or other appropriate medium including at least one of memory 1002 and storage 1003.

[0059] The communication device 1004 is hardware (transmission / reception device) for communicating between computers via at least one of a wired network and a wireless network, and is also referred to as, for example, a network device, a network controller, a network card, or a communication module. The communication device 1004 may be configured to include a high-frequency switch, a duplexer, a filter, a frequency synthesizer, etc. to realize at least one of frequency division duplex (FDD) and time division duplex (TDD). For example, the above-mentioned data receiving unit 11, second estimating unit 13, etc. may be realized by the communication device 1004.

[0060] The input device 1005 is an input device (for example, a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that receives input from the outside. The output device 1006 is an output device (for example, a display, a speaker, an LED lamp, etc.) that outputs to the outside. The input device 1005 and the output device 1006 may be integrated into one device (for example, a touch panel).

[0061] Furthermore, each device, such as the processor 1001 and the memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or may be configured using different buses between each device.

[0062] The estimation device 1 may also be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA), and some or all of the functional blocks may be realized by the hardware. For example, the processor 1001 may be implemented using at least one of these pieces of hardware.

[0063] The notification of information is not limited to the aspects / embodiments described in the present disclosure and may be performed using other methods. For example, the notification of information may be performed by physical layer signaling (e.g., Downlink Control Information (DCI), Uplink Control Information (UCI)), higher layer signaling (e.g., Radio Resource Control (RRC) signaling, Medium Access Control (MAC) signaling, broadcast information (Master Information Block (MIB), System Information Block (SIB))), other signals, or a combination thereof. Furthermore, the RRC signaling may be referred to as an RRC message, and may be, for example, an RRC Connection Setup message, an RRC Connection Reconfiguration message, or the like.

[0064] The order of the procedures, sequences, flowcharts, etc. of each aspect / embodiment described in this disclosure may be changed unless it is consistent. For example, the methods described in this disclosure present elements of various steps using an example order, and are not limited to the particular order presented.

[0065] Input and output information may be stored in a specific location (for example, memory) or may be managed using a management table. Input and output information may be overwritten, updated, or added to. Output information may be deleted. Input information may be sent to another device.

[0066] The determination may be made based on a value represented by one bit (0 or 1), a Boolean value (true or false), or a numerical comparison (e.g., comparison with a predetermined value).

[0067] Each aspect / embodiment described in this disclosure may be used alone, in combination, or switched depending on the implementation. Furthermore, notification of predetermined information (e.g., notification that "X is true") is not limited to being done explicitly, but may be done implicitly (e.g., by not notifying the predetermined information).

[0068] Although the present disclosure has been described in detail above, it is clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the spirit and scope of the present disclosure as defined by the claims. Therefore, the description of the present disclosure is intended to be illustrative and does not have any limiting meaning on the present disclosure.

[0069] Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0070] Software, instructions, information, etc. may also be transmitted or received over a transmission medium. For example, if software is transmitted from a website, server, or other remote source using wired technologies (such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)), and / or wireless technologies (such as infrared, microwave), these wired and / or wireless technologies are included within the definition of transmission media.

[0071] The information, signals, etc. described in this disclosure may be represented using any of a variety of different technologies. For example, data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0072] Note that terms explained in this disclosure and terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings. For example, at least one of a channel and a symbol may be a signal (signaling). Furthermore, a signal may be a message. Furthermore, a component carrier (CC) may be called a carrier frequency, a cell, a frequency carrier, etc.

[0073] Furthermore, the information, parameters, etc. described in the present disclosure may be expressed using absolute values, may be expressed using relative values ​​from a predetermined value, or may be expressed using other corresponding information. For example, a radio resource may be indicated by an index.

[0074] The names used for the above-described parameters are not intended to be limiting in any way. Furthermore, the mathematical expressions using these parameters may differ from those explicitly disclosed in this disclosure. The various channels (e.g., PUCCH, PDCCH, etc.) and information elements may be identified by any suitable names, and therefore the various names assigned to these various channels and information elements are not intended to be limiting in any way.

[0075] In this disclosure, the terms "Mobile Station (MS)," "user terminal," "User Equipment (UE)," "terminal," etc. may be used interchangeably.

[0076] A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or some other suitable terminology.

[0077] As used in this disclosure, the terms "determining" and "determining" may encompass a wide variety of actions. "Determining" and "determining" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, inquiring (e.g., searching in a table, database, or other data structure), ascertaining, and the like. "Determining" and "determining" may also include receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, accessing (e.g., accessing data in memory), and the like. Furthermore, "judgment" and "decision" can include regarding resolving, selecting, choosing, establishing, comparing, etc. as having been "judged" or "decided." In other words, "judgment" and "decision" can include regarding some action as having been "judged" or "decided." Furthermore, "judgment (decision)" can be interpreted as "assuming," "expecting," "considering," etc.

[0078] The terms "connected," "coupled," or any variation thereof, refer to any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are "connected" or "coupled" to each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, "connected" may be read as "access." As used in this disclosure, two elements may be considered to be "connected" or "coupled" to each other using one or more wires, cables, and / or printed electrical connections, as well as electromagnetic energy having wavelengths in the radio frequency range, microwave range, and optical (both visible and invisible) range, as some non-limiting and non-exhaustive examples.

[0079] As used in this disclosure, the phrase "based on" does not mean "based only on," unless expressly stated otherwise. In other words, the phrase "based on" means both "based only on" and "based at least on."

[0080] Any reference to an element using a designation such as "first," "second," etc., used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient way to distinguish between two or more elements. Thus, a reference to a first and a second element does not imply that only two elements may be employed or that the first element must in some way precede the second element.

[0081] When used in this disclosure, the terms "include," "including," and variations thereof are intended to be inclusive, similar to the term "comprising." Furthermore, when used in this disclosure, the term "or" is not intended to be an exclusive or.

[0082] In this disclosure, where articles are added by translation, such as a, an, and the in English, the disclosure may include that the nouns following these articles are in the plural form.

[0083] In the present disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "coupled" may also be interpreted in the same way as "different." [Explanation of symbols]

[0084] 1...estimation device, 3a, 3b, 3c...Wi-Fi receiver (short-range wireless communication device), 11...data receiving unit, 13...second estimation unit, 14...data generating unit, 15...training unit, TDA...training data for unsupervised learning (signal state data), TD1...first training data, TD2...second training data, p i m …pseudo-label, q i m ,y i m ...correct value, L...additive loss value (value of the additive loss function), L1...loss value (value of the first loss function), L2...loss value (value of the second loss function).

Claims

1. a data receiving unit that receives signal state data relating to the amplitude or phase of a received signal from the short-range wireless communication device in time series; an estimation unit that inputs input data based on the time-series signal state data into a learning model and estimates human behavior based on an output of the learning model; a data generating unit that performs first data extension on the time-series signal state data received by the data receiving unit to generate first training data, and performs second data extension on the time-series signal state data to generate second training data; a training unit that assigns pseudo labels to the first training data based on a result of inputting the first training data into the learning model, and trains the learning model using the second training data and the pseudo labels; An estimation device comprising:

2. The first data extension is a process of weakly transforming the time-series signal condition data, The second data extension is a process of strongly transforming the time-series signal condition data. The estimation device according to claim 1 .

3. the training unit calculates a first loss function based on a result of inputting the second training data into the learning model and the pseudo label, and updates parameters of the learning model using the first loss function. The estimation device according to claim 1 .

4. the training unit further acquires third training data in which correct values ​​are labeled with respect to the time-series signal state data, further calculates a second loss function based on a result of inputting the third training data into the learning model and the correct values, and further uses the second loss function to update parameters of the learning model. The estimation device according to claim 3 .

5. the training unit calculates an additive loss function by adding the value of the first loss function and the value of the second loss function, and updates parameters of the learning model based on the additive loss function. The estimation device according to claim 4 .

6. the training unit calculates an additive loss function by performing a weighted addition of the value of the first loss function and the value of the second loss function, with the weighting changed according to the number of epochs, and updates parameters of the learning model based on the additive loss function; The estimation device according to claim 4 .

7. The first data extension is a process of extracting two pairs of time-series signal state data from a plurality of time-series signal state data, and transforming one of the two time-series signal state data so that it also has characteristics of the other of the two time-series signal state data. The estimation device according to claim 2 .

8. the second data expansion is a spline interpolation process; The estimation device according to claim 2 .

Images