Learning method
The learning method addresses performance deterioration in time-series data feature extraction by updating the model to align features within the same data characteristic, enhancing extraction accuracy and adaptability across different data conditions.
Patent Information
- Application Number
- PCT/JP2023/019312
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-24
- Publication Date
- 2024-11-28
AI Technical Summary
Existing learning-based methods for converting time-series data into feature vectors face performance deterioration due to changes in data characteristics between model learning and retrieval, especially when data is collected using different devices, making it challenging to maintain effective feature extraction.
A learning method that acquires multiple time-series datasets with different data characteristics, extracts features, and updates the model by comparing feature quantities for each data characteristic, using supervised distance learning to separate and align features within the same data characteristic, thereby improving feature extraction performance.
This approach enhances the accuracy of feature extraction and classification by adapting the model to varying data characteristics, ensuring more appropriate feature representation and improved performance even when data characteristics differ between learning and search phases.
Smart Images

Figure JP2023019312_28112024_PF_FP_ABST
Abstract
Description
Learning methods
[0001] The present invention relates to a learning method, a learning device, and a recording medium.
[0002] Techniques used to detect anomalies based on time-series data are known.
[0003] For example, Patent Document 1 describes a time series data processing device that converts a time series data set, which is a collection of multiple time series data, into a feature vector that indicates characteristics of the time series data set, thereby enabling search. Specifically, Patent Document 1 describes a time series data processing device that includes a data conversion unit, a storage unit, and a search unit. For example, the data conversion unit converts partial time series data sets, which are obtained by dividing a time series data set, which is a collection of multiple time series data, into feature vectors that indicate characteristics of the partial time series data sets. The storage unit also stores multiple first partial time series data sets and multiple first feature vectors converted from the multiple first partial time series data sets by the data conversion unit, in association with each other. The search unit then selects at least one first feature vector from the multiple first feature vectors stored in the storage unit that is similar to a second feature vector converted from an input second partial time series data set by the data conversion unit, and outputs the first partial time series data set corresponding to the selected first feature vector.
[0004] International Publication No. 2020 / 049666
[0005] When converting a time series data set into a feature vector (feature values), the conversion may be achieved using a pre-trained model. When using such a learning-based method, it is generally assumed that data characteristics, such as data distribution, do not change between model training and search. However, in reality, data characteristics may differ between training and search. As a result, there is a risk of performance degradation during feature conversion (extraction). Furthermore, even during training, for example, when time series data is collected using different devices, data characteristics may differ between the devices. As a result, time series data with different data characteristics may be integrated and trained, which may result in performance degradation during feature extraction, as in the above case. Thus, when training a model that converts a time series data set into a feature vector (feature values), it may be difficult to train a model that suppresses performance degradation.
[0006] Therefore, one of the objects of the present invention is to provide a learning method, a learning device, and a recording medium that can solve the above-mentioned problems.
[0007] In order to achieve this object, a learning method that is one aspect of the present disclosure is configured such that an information processing device acquires a plurality of time-series data sets with different data characteristics, extracts features based on the acquired time-series data sets, and updates a model used to extract the features by comparing the features for each data characteristic of the source of feature extraction based on the extracted features.
[0008] Furthermore, a learning device according to another aspect of the present disclosure includes a time-series data acquisition unit that acquires a plurality of time-series data sets with different data characteristics; a feature extraction unit that extracts features based on the time-series data sets acquired by the time-series data acquisition unit; and a model learning unit that updates a model used when extracting the features by comparing the features extracted by the feature extraction unit for each data characteristic of the source of feature extraction.
[0009] Furthermore, a recording medium as another aspect of the present disclosure is a computer-readable recording medium having recorded thereon a program for causing an information processing device to perform the following processes: acquire a plurality of time-series data sets having different data characteristics; extract features based on the acquired time-series data sets; and update a model used for extracting the features by comparing the features for each data characteristic of the source of feature extraction based on the extracted features.
[0010] According to the above-mentioned configurations, the above-mentioned problems can be solved.
[0011] FIG. 1 is a diagram illustrating an overview of a learning system according to a first embodiment of the present disclosure. FIG. 2 is a diagram illustrating an overview of the learning system. FIG. 3 is a diagram illustrating an overview of the learning system. FIG. 4 is a block diagram illustrating an example configuration of a learning device. FIG. 5 is a diagram illustrating an example of a learning process. FIG. 6 is a flowchart illustrating an example operation of a learning device. FIG. 7 is a diagram illustrating an example of an effect of the present disclosure. FIG. 8 is a block diagram illustrating another example configuration of a learning device. FIG. 9 is a diagram illustrating an example hardware configuration of a learning device according to a second embodiment of the present disclosure. FIG. 10 is a block diagram illustrating an example configuration of a learning device.
[0012] [First Embodiment] A first embodiment of the present invention will be described with reference to Figs. 1 to 8. Figs. 1 to 3 are diagrams for explaining an overview of a learning system 100. Fig. 4 is a block diagram showing an example of the configuration of a learning device 200. Fig. 5 is a diagram for explaining an example of a learning process performed by the learning device 200. Fig. 6 is a flowchart showing an example of the operation of the learning device 200. Fig. 7 is a diagram for explaining an example of the effect of the present disclosure. Fig. 8 is a block diagram showing another example of the configuration of the learning device 200.
[0013] In a first embodiment of the present disclosure, a learning system 100 is described that learns or updates a model based on a time-series dataset consisting of one or more time-series data acquired using one or more devices. For example, as shown in FIG. 1 , the learning system 100 acquires time-series datasets with different data characteristics. The learning system 100 also extracts features from each acquired time-series dataset. For example, the learning system 100 extracts features by inputting time-series segments, which are obtained by dividing the time-series dataset into predetermined time intervals, into a pre-trained model. The learning system 100 then updates the model used for feature extraction by comparing features for each data characteristic of the time-series dataset from which the features were extracted. In other words, the learning system 100 updates the model used for feature extraction by comparing, among multiple features, features that have the same data characteristics as the source of extraction. In this way, the learning system 100 separates features according to the source data characteristics and performs metric learning on each feature.
[0014] In the present disclosure, data characteristics refer to the data distribution in a time series dataset. For example, data characteristics may vary depending on, for example, changing at least some of the devices, such as sensors, that acquire each piece of time series data included in the time series dataset. Furthermore, when extracting features for purposes such as fault classification of an optical transponder, time series data may be acquired using devices included in the optical transponder. In such cases, data characteristics may also vary when an optical cable connected to the optical transponder is unplugged or plugged in. The learning system 100 in the present disclosure acquires time series datasets with different data characteristics by changing the acquisition status of the time series data, such as by changing the device that acquires the time series data or by unplugging and plugging in an optical cable connected to the device that acquires the time series data. Time series datasets with different data characteristics may also be acquired using methods other than those exemplified above.
[0015] Furthermore, the learning system 100 described in this disclosure can train a model using a supervised distance learning method such as that described in Patent Document 1. For example, the learning system 100 randomly extracts anchor samples from a group of features consisting of multiple features extracted from a time-series dataset, and extracts positive samples assigned to the same class as the anchor samples and negative samples assigned to a different class from the anchor samples. Then, as illustrated in FIG. 2 , the learning system 100 trains and updates the model by updating weight parameters so as to reduce the distance between the anchor samples and positive samples and increase the distance between the anchor samples and negative samples in the feature space.
[0016] In the present disclosure, the feature set includes features extracted from time-series data sets with different data characteristics. Therefore, as illustrated in FIG. 3 , simply focusing on the classes assigned to the features may result in extracting a feature as a positive sample that is assigned the same class as the anchor sample but has different source data characteristics. However, such features have different distributions in the original data and should not be approximated in feature space. Therefore, the learning system 100 described in the present disclosure extracts anchor samples, positive samples, and negative samples from among the features included in the feature set that have the same source data characteristics. In this way, the learning system 100 can compare features within the same data characteristics and perform model updating, thereby training a model that enables more appropriate extraction.
[0017] FIG. 1 shows an overview of a learning system 100. Referring to FIG. 1, the learning system 100 includes a learning device 200, which is an information processing device that learns or updates a model based on multiple time-series data sets with different data characteristics. As shown in FIG. 1, the learning device 200 acquires time-series data sets by acquiring time-series data from various sensors and other devices. Furthermore, the learning device 200 learns or updates a model based on a time-series data set consisting of one or more pieces of time-series data.
[0018] Fig. 4 shows an example configuration of learning device 200. Referring to Fig. 4, learning device 200 has, as main components, for example, an operation input unit 210, a screen display unit 220, a communication I / F unit 230, a storage unit 240, and an arithmetic processing unit 250.
[0019] 4 illustrates an example in which the functions of learning device 200 are realized using a single information processing device. However, at least some of the functions of learning device 200 may be realized using multiple information processing devices, for example, on the cloud. Learning device 200 may not include some of the components illustrated above, such as not having operation input unit 210 or screen display unit 220, or may have components other than those illustrated above.
[0020] Operation input unit 210 is made up of operation input devices such as a keyboard, a mouse, etc. Operation input unit 210 detects operations of the operator operating learning device 200 and outputs the operations to calculation processing unit 250.
[0021] The screen display unit 220 is composed of a screen display device such as a liquid crystal display, an organic electroluminescence (EL) display, etc. The screen display unit 220 can display various information stored in the storage unit 240 on the screen in response to instructions from the arithmetic processing unit 250.
[0022] The communication I / F unit 230 includes a data communication circuit, etc. The communication I / F unit 230 performs data communication with various sensors and other external devices connected via communication lines.
[0023] The storage unit 240 is a storage device such as a hard disk or memory. The storage unit 240 stores processing information and a program 244 required for various processes in the arithmetic processing unit 250. The program 244 is read into the arithmetic processing unit 250 and executed to realize various processing units. The program 244 is read in advance from an external device or recording medium via a data input / output function such as the communication I / F unit 230, and is stored in the storage unit 240. Main information stored in the storage unit 240 includes, for example, model information 241, feature amount information 242, and time-series data information 243.
[0024] The model information 241 includes information about a model that extracts and outputs features that preserve local distance relationships for input time-series segments. For example, the model information 241 may include weight parameters included in the trained model described above. For example, the model included in the model information 241 is trained in advance using training time-series segments inside or outside the training device 200 and stored in the storage unit 240. The model information 241 is updated through training by the model training unit 254, which will be described later.
[0025] The feature information 242 includes information corresponding to features extracted from the time-series segments. For example, the feature information 242 includes binary codes obtained by converting features using any method. In the feature information 242, the binary codes may be associated with classes to be classified. For example, the binary codes included in the feature information 242 are obtained by converting features extracted by the feature extraction unit 252 using the binary code conversion unit 253, and are stored in the storage unit 240.
[0026] The feature amount information included in the feature amount information 242 can be used when performing a search process, which will be described later. For example, the results of the search process may be used when determining whether or not an abnormality exists or to what class the data belongs. The results of the search process may also be used for purposes other than those exemplified above.
[0027] The time-series data information 243 includes a time-series data set consisting of one or more pieces of time-series data. Here, the time-series data may be data in which numerical data, such as observation data measured by a sensor at a predetermined interval, is arranged in order of measurement time. For example, the time-series data information 243 includes a time-series data set for each data characteristic. The time-series data information 243 is updated in response to the time-series data acquisition unit 251 (described later) acquiring time-series data from sensors and other devices.
[0028] The arithmetic processing unit 250 has an arithmetic device such as a CPU (Central Processing Unit) and its peripheral circuits. The arithmetic processing unit 250 reads and executes a program 244 from the storage unit 240, thereby realizing various processing functions by causing the above hardware and the program 244 to work together. Major processing units realized by the arithmetic processing unit 250 include, for example, a time-series data acquisition unit 251, a feature extraction unit 252, a binary code conversion unit 253, and a model learning unit 254.
[0029] In addition, the arithmetic processing unit 250 may have a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating point number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination of these, instead of the above-mentioned CPU.
[0030] The time-series data acquisition unit 251 acquires a time-series data set by acquiring time-series data from various sensors and other devices included in the learning system 100. For example, the time-series data acquisition unit 251 may acquire time-series data from an optical transponder, an optical performance monitor that measures the optical signal to noise ratio (OSNR), multiple sensors installed in a plant or data center, or any other device.
[0031] Furthermore, the time-series data acquisition unit 251 can acquire time-series data sets with different data characteristics. The time-series data acquisition unit 251 may acquire multiple time-series data sets with different data characteristics. For example, the time-series data acquisition unit 251 can acquire time-series data sets with different data characteristics by acquiring time-series data sets while changing the acquisition status of the time-series data at predetermined time intervals. Note that the acquisition status of the time-series data may be changed automatically by executing a predetermined process, or may be changed manually.
[0032] The time-series data acquisition unit 251 also stores the acquired time-series data set in the storage unit 240 as time-series data information 243. At this time, the time-series data acquisition unit 251 can store the time-series data set for each data characteristic in the storage unit 240. The time-series data acquisition unit 251 may determine the data characteristics of the time-series data set using any method. For example, the time-series data acquisition unit 251 may determine the data characteristics of the time-series data set by acquiring information indicating the acquisition status of the time-series data. The time-series data acquisition unit 251 may acquire information indicating the acquisition status of the time-series data in response to an operation by an operator on the operation input unit 210, or may acquire information indicating the acquisition status of the time-series data from a device that acquires the time-series data via the communication I / F unit 230. The time-series data acquisition unit 251 may also detect a fluctuation in the data characteristics in response to a result of detecting a fluctuation in the distribution of the time-series data using, for example, any statistical method.
[0033] The feature extraction unit 252 extracts features from the time series segments by inputting the time series segments to a model stored as the model information 241. For example, the feature extraction unit 252 divides the time series data set acquired by the time series data acquisition unit 251 into multiple time series segments using a time window of a certain period. Then, the feature extraction unit 252 inputs each of the divided time series segments to a trained model to extract features. Note that the feature extraction unit 252 may divide the time series data into the time series segments using any method. For example, the size of the time window may be set arbitrarily. Furthermore, the feature extraction unit 252 may divide the time series data into multiple time series segments so that they overlap for an arbitrary period, or may divide the time series data into multiple time series segments so that the time series segments do not overlap.
[0034] The binary code conversion unit 253 converts the features extracted by the feature extraction unit 252 into binary code, which is information corresponding to the features. In the present disclosure, the method of conversion into binary code is not particularly limited. The binary code conversion unit 253 may convert the features extracted by the feature extraction unit 252 into binary code using any method. Furthermore, the binary code conversion unit 253 can store the converted binary code in the storage unit 240 as feature information 242.
[0035] The feature extraction unit 252 or the binary code conversion unit 253 may assign a class to be classified to the feature or binary code using any method. For example, the feature extraction unit 252 or the binary code conversion unit 253 may assign a class to the feature, etc., determined based on the Euclidean distance between the time-series segment from which the feature was extracted and the time-series segment to be compared. Information about the class may be acquired based on an operation by the operator on the operation input unit 210, etc.
[0036] The model learning unit 254 performs learning based on the features extracted by the feature extraction unit 252, thereby updating the model stored as model information 241. The model learning unit 254 can perform the above learning by comparing the features for each data characteristic of the extraction source.
[0037] For example, the model training unit 254 trains the model using a supervised metric learning method such as that described in Patent Document 1. The model training unit 254 can train the model so that metric learning loss within the same data characteristics is minimized. As an example, the model training unit 254 randomly extracts anchor samples from a feature group including multiple feature values extracted by the feature extraction unit 252, and extracts positive samples assigned to the same class as the anchor sample and negative samples assigned to a different class from the anchor sample. In this case, the model training unit 254 extracts anchor samples, positive samples, and negative samples from feature values included in the feature group that have the same data characteristics as the original samples. The model training unit 254 then updates the model by comparing the extracted samples. For example, the model training unit 254 can update the model by updating weight parameters so as to shorten the distance between the anchor sample and the positive sample and widen the distance between the anchor sample and the negative sample in the feature space.
[0038] FIG. 5 shows an example of the learning process performed by the model learning unit 254 described above. As shown in FIG. 5, by extracting each sample from feature quantities having the same original data characteristics, the model learning unit 254 can compare each feature quantity within the same data characteristics. For example, in the example shown in FIG. 5, the model learning unit 254 performs a comparison within data characteristics A, a comparison within data characteristics B, and a comparison within data characteristics C. The model learning unit 254 may repeat the extraction and comparison of each sample within the same data characteristics until a predetermined condition is satisfied for each data characteristic. Note that the number of data characteristics, etc. may be different from that shown in FIG. 5.
[0039] The class corresponding to each feature may be assigned by the feature extraction unit 252 or the binary code conversion unit 253, or may be assigned by the model learning unit 254. For example, the model learning unit 254 may assign a class to each feature using a method similar to that used by the feature extraction unit 252 or the binary code conversion unit 253.
[0040] The loss function used by the model training unit 254 during training may be any function that preserves local similarity between data in the input space. For example, the loss function may be triplet loss, pairwise loss, contrastive loss, or the like. The model trained by the model training unit 254 may be any model that can handle time-series data. For example, the model may be one of a 1D-CNN (1-Dimensional-Convolutional Neural Network), a GRU (Gated Recurrent Unit), a LSTM (Long Short-Term Memory), a Transformer, or the like.
[0041] The above is an example of the configuration of the learning device 200. Next, an example of the operation of the learning device 200 will be described with reference to FIG.
[0042] 6 shows an example of the operation of the learning device 200. Referring to Fig. 6, the time-series data acquisition unit 251 acquires a time-series data set by acquiring time-series data from various sensors and other devices included in the learning system 100. For example, the time-series data acquisition unit 251 can acquire multiple time-series data sets with different data characteristics (step S101).
[0043] The feature extraction unit 252 extracts features based on the time series segments by inputting the time series segments to the model stored as the model information 241 (step S102). For example, the feature extraction unit 252 divides the time series data set acquired by the time series data acquisition unit 251 into multiple time series segments using a time window of a certain period. Then, the feature extraction unit 252 inputs each divided time series segment to the trained model to extract features.
[0044] The model learning unit 254 performs learning based on the features extracted by the feature extraction unit 252, thereby updating the model stored as the model information 241. The model learning unit 254 can perform the above learning by comparing the features for each data characteristic of the extraction source (step S103).
[0045] For example, the model learning unit 254 randomly extracts anchor samples from a feature group containing multiple feature values extracted by the feature extraction unit 252, and also extracts positive samples assigned to the same class as the anchor sample and negative samples assigned to a different class from the anchor sample. In this case, the model learning unit 254 extracts anchor samples, positive samples, and negative samples from feature values included in the feature group that have the same data characteristics as the original samples. The model learning unit 254 then updates the model by comparing the extracted samples. For example, the model learning unit 254 can update the model by updating weight parameters so as to reduce the distance between the anchor sample and the positive sample and increase the distance between the anchor sample and the negative sample in the feature space.
[0046] As described above, the learning device 200 includes a time-series data acquisition unit 251, a feature extraction unit 252, and a model learning unit 254. With this configuration, the feature extraction unit 252 can extract features based on time-series data sets with different data characteristics acquired by the time-series data acquisition unit 251. Furthermore, the model learning unit 254 can update the model by comparing features extracted by the feature extraction unit 252 for each data characteristic of the source data based on the features extracted from the time-series data sets with different data characteristics. As a result, learning can be performed taking into account differences in data characteristics, enabling appropriate learning. Furthermore, by learning various data characteristics in advance, more appropriate feature extraction can be expected even when data characteristics differ between learning and search.
[0047] FIG. 7 shows an example of the effect of learning by the learning device 200 described in the present disclosure. For example, in the example shown in FIG. 7, the x-axis indicates the class predicted when a prediction is made based on features extracted by the model learned by the model learning unit 254, and the y-axis indicates the actual correct class. Furthermore, in the example shown in FIG. 7, a correct prediction is indicated when the prediction is located on a diagonal line from the top left to the bottom right. Referring to FIG. 7, it can be seen that the accuracy rate is improved by learning using the learning device 200 described in the present disclosure, and more accurate predictions are possible. In other words, it can be seen that the configuration described in the present disclosure allows for more appropriate feature extraction to achieve accurate predictions.
[0048] The configuration of the learning device 200 is not limited to the configuration exemplified with reference to Fig. 4. For example, the learning device 200 may be configured with a portion of the configuration exemplified in Fig. 4, such as omitting the binary code conversion unit 253. Fig. 8 also shows another example configuration of the learning device 200. Referring to Fig. 8, for example, the arithmetic processing unit 250 can implement a search unit 255 in addition to the configuration exemplified in Fig. 4 by reading and executing the program 244 from the storage unit 240.
[0049] The search unit 255 performs search processing based on the feature amount extracted from the time-series segment to be searched.
[0050] For example, when the time-series data acquisition unit 251 acquires a time-series data set to be searched, the feature extraction unit 252 divides the time-series data set to be searched into multiple time-series segments and extracts features. Furthermore, the binary code conversion unit 253 converts the features extracted by the feature extraction unit 252 into binary codes, which are information corresponding to the features. The search unit 255 acquires the binary codes converted as described above. The search unit 255 then searches the feature information 242 for binary codes similar to the acquired binary code. For example, the search unit 255 calculates the distance between the acquired binary code and each binary code included in the feature information 242. The search unit 255 then acquires, as binary codes similar to the acquired binary code, a binary code that satisfies an arbitrary condition, such as a binary code with the smallest calculated distance, from among the binary codes included in the feature information 242.
[0051] The search unit 255 can also perform various processes according to the search results. For example, the search unit 255 can infer the class of the time-series segment to be searched by identifying a class associated with the binary code acquired by the search. The search unit 255 can also detect an anomaly according to the search results. For example, the search unit 255 calculates an anomaly score according to the distance between the binary code to be searched and the searched binary code. Then, the search unit 255 can detect an anomaly based on the calculated anomaly score. For example, the search unit 255 can detect an anomaly based on a comparison result between the calculated anomaly score and a predetermined threshold. The search unit 255 can also perform processes according to search results other than those exemplified above.
[0052] Second Embodiment Next, a second embodiment of the present disclosure will be described with reference to Fig. 9 and Fig. 10. Fig. 9 is a diagram illustrating an example of the hardware configuration of a learning device 300. Fig. 10 is a block diagram illustrating an example of the configuration of the learning device 300.
[0053] In a second embodiment of the present disclosure, a learning device 300 will be described, which is an information processing device that updates a model based on time-series data sets with different data characteristics. FIG. 9 shows an example hardware configuration of the learning device 300. Referring to FIG. 9 , the learning device 300 has, as an example, the following hardware configuration: CPU (Central Processing Unit) 301 (arithmetic device); ROM (Read Only Memory) 302 (storage device); RAM (Random Access Memory) 303 (storage device); programs 304 loaded into RAM 303; a storage device 305 that stores the programs 304; a drive device 306 that reads and writes data from and to a storage medium 310 external to the information processing device; a communication interface 307 that connects to a communication network 311 external to the information processing device; an input / output interface 308 that inputs and outputs data; and a bus 309 that connects the various components.
[0054] 10 . The program group 304 is stored in advance in the storage device 305 or the ROM 302, for example, and is loaded into the RAM 303 or the like and executed by the CPU 301 as needed. The program group 304 may be supplied to the CPU 301 via the communication network 311, or may be stored in advance in the recording medium 310, with the drive device 306 reading out the program and supplying it to the CPU 301.
[0055] 9 shows an example of the hardware configuration of the learning device 300. The hardware configuration of the learning device 300 is not limited to the above. For example, the learning device 300 may be configured with only a part of the above configuration, such as not including the drive device 306. Furthermore, the CPU 301 may be a GPU, as exemplified in the first embodiment.
[0056] The time-series data acquisition unit 321 acquires a plurality of time-series data sets having different data characteristics. For example, a time-series data set includes one or more pieces of time-series data.
[0057] The feature extraction unit 322 extracts features based on the time-series data set acquired by the time-series data acquisition unit 321. For example, the feature extraction unit 322 extracts features by inputting time-series segments obtained by dividing the time-series data set using a predetermined time window into a pre-trained model.
[0058] The model learning unit 323 updates the model used when extracting the features by comparing the features with each data characteristic of the feature extraction source based on the features extracted by the feature extraction unit 322 .
[0059] As described above, the learning device 300 includes the model learning unit 323. With this configuration, the model learning unit 323 can update the model used to extract features by comparing features for each data characteristic from which the features were extracted, based on the features extracted by the feature extraction unit 322. As a result, learning can be performed that takes differences in data characteristics into account, enabling appropriate learning.
[0060] The above-described learning device 300 can be realized by incorporating a predetermined program into an information processing device such as the learning device 300. Specifically, a program according to another aspect of the present invention is a program for causing an information processing device such as the learning device 300 to acquire multiple time-series data sets with different data characteristics, extract features based on the acquired time-series data sets, and, based on the extracted features, compare the features for each data characteristic from which the features were extracted, thereby updating a model used when extracting the features.
[0061] Furthermore, a learning method executed by an information processing device such as the above-described learning device 300 is a method in which the information processing device acquires multiple time-series data sets with different data characteristics, extracts features based on the acquired time-series data sets, and updates a model used to extract features by comparing features for each data characteristic from which the features were extracted based on the extracted features.
[0062] Even if the invention is a program having the above-mentioned configuration, or a computer-readable recording medium on which the program is recorded, or a learning method, it can achieve the same functions and effects as the above-mentioned learning device 300, and therefore can achieve the above-mentioned objective of the present disclosure.
[0063] <Supplementary Notes> Part or all of the above-described embodiments can be described as follows: The learning method and the like of the present invention will be outlined below. However, the present invention is not limited to the following configuration.
[0064] (Supplementary Note 1) A learning method in which an information processing device acquires multiple time-series datasets with different data characteristics, extracts features based on the acquired time-series datasets, and updates a model used to extract the features based on the extracted features by comparing the features for each data characteristic from which the features were extracted. (Supplementary Note 2) The learning method according to Supplementary Note 1, wherein, when acquiring the time-series datasets, the time-series datasets with different data characteristics are acquired by acquiring the time-series datasets while changing the time-series data acquisition status. (Supplementary Note 3) The learning method according to Supplementary Note 1 or Supplementary Note 2, wherein, when acquiring the time-series datasets, the time-series datasets with different data characteristics are acquired by plugging and unplugging an optical cable connecting to a device. (Supplementary Note 4) The learning method according to any one of Supplements 1 to 3, wherein, when acquiring the time-series datasets, time-series datasets with different data distributions are acquired as the multiple time-series datasets with different data characteristics. (Supplementary Note 5) The learning method according to any one of Supplementary Notes 1 to 4, wherein, when updating a model, a model used when extracting features is updated by extracting a plurality of feature amounts from the feature amounts having the same data characteristics as the extraction source and comparing the extracted feature amounts. (Supplementary Note 6) The learning method according to any one of Supplementary Notes 1 to 5, wherein, when updating a model, an anchor sample, a positive sample assigned the same class as the anchor sample, and a negative sample assigned a different class from the anchor sample are extracted from the feature amounts having the same data characteristics as the extraction source, and the extracted anchor sample, the positive sample, and the negative sample are compared. (Supplementary Note 7) The learning method according to any one of Supplementary Notes 1 to 6, wherein, when updating a model, a weight parameter included in the model is updated by comparing each data characteristic of the feature amount extraction source.(Supplementary Note 8) The learning method according to any one of Supplementary Notes 1 to 7, wherein when updating the model, the model is updated so that distance learning loss within the same data characteristics is minimized. (Supplementary Note 9) A learning device comprising: a time-series data acquisition unit that acquires multiple time-series data sets with different data characteristics; a feature extraction unit that extracts features based on the time-series data sets acquired by the time-series data acquisition unit; and a model learning unit that updates a model used when extracting the features by comparing the features extracted by the feature extraction unit for each data characteristic of a source of feature extraction. (Supplementary Note 10) A computer-readable recording medium having recorded thereon a program for implementing in an information processing device the following processes: acquiring multiple time-series data sets with different data characteristics, extracting features based on the acquired time-series data sets, and updating a model used when extracting the features by comparing the features for each data characteristic of the source of feature extraction based on the extracted features.
[0065] The programs described in the above embodiments and appendices may be stored in a storage device or a computer-readable recording medium, such as a portable medium such as a flexible disk, an optical disk, a magneto-optical disk, or a semiconductor memory.
[0066] Although the present invention has been described above with reference to the above-mentioned embodiments, the present invention is not limited to the above-mentioned embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.
[0067] 100 Learning system 200 Learning device 210 Operation input unit 220 Screen display unit 230 Communication I / F unit 240 Storage unit 241 Model information 242 Feature information 243 Time series data information 244 Program 250 Arithmetic processing unit 251 Time series data acquisition unit 252 Feature extraction unit 253 Binary code conversion unit 254 Model learning unit 255 Search unit 300 Learning device 301 CPU 302 ROM 303 RAM 304 Program group 305 Storage device 306 Drive device 307 Communication interface 308 Input / output interface 309 Bus 310 Recording medium 311 Communication network 321 Time series data acquisition unit 322 Feature extraction unit 323 Model learning unit
Claims
1. A learning method in which an information processing device acquires multiple time-series data sets with different data characteristics, extracts features based on the acquired time-series data sets, and updates a model used when extracting the features by comparing the features for each data characteristic from which the features were extracted based on the extracted features.
2. The learning method according to claim 1, wherein, when acquiring a time-series data set, a time-series data set with different data characteristics is acquired by acquiring the time-series data set while changing the acquisition conditions of the time-series data.
3. A learning method according to claim 1, wherein, when acquiring a time series data set, an optical cable connected to the device is unplugged and plugged in to acquire a time series data set having different data characteristics.
4. A learning method according to claim 1, wherein when acquiring a time series data set, time series data sets with different data distributions are acquired as the multiple time series data sets with different data characteristics.
5. A learning method as claimed in claim 1, wherein when updating a model, a plurality of features are extracted from the features having the same data characteristics from which the features were extracted, and a comparison is made between the extracted features, thereby updating the model used when extracting the features.
6. A learning method as described in claim 1, wherein, when updating a model, an anchor sample, a positive sample assigned the same class as the anchor sample, and a negative sample assigned a different class from the anchor sample are extracted from among the features that have the same data characteristics as the features from which they were extracted, and the extracted anchor sample, positive sample, and negative sample are compared.
7. A learning method according to claim 1, wherein when updating the model, weight parameters included in the model are updated by comparison with each data characteristic from which the features are extracted, thereby updating the model.
8. The learning method according to claim 1, wherein, when updating the model, the model is updated so that a distance learning loss within the same data characteristics is minimized.
9. A learning device comprising: a time series data acquisition unit that acquires a plurality of time series data sets having different data characteristics; a feature extraction unit that extracts features based on the time series data sets acquired by the time series data acquisition unit; and a model learning unit that updates a model used when extracting the features by comparing the features for each data characteristic from which the features were extracted based on the features extracted by the feature extraction unit.
10. A computer-readable recording medium having recorded thereon a program for implementing a process in an information processing device, which comprises: acquiring multiple time-series data sets having different data characteristics; extracting features based on the acquired time-series data sets; and updating a model used in extracting the features by comparing the features for each data characteristic from which the features were extracted based on the extracted features.
Citation Information
Patent Citations
Estimation method, estimation device, and program
WO2021224959A1
Information processing device, information processing method, and information processing program
WO2022038660A1
