Learning apparatus

Abstract

To solve the problem in which it is difficult to realize efficient machine learning in response to a change in input dimension.SOLUTION: A learning apparatus includes: an extraction unit that extracts a feature value from a time-series data set including time-series data on predetermined observation items by using a machine-learned model; a detection unit that detects a change in an input dimension, which is a change in the number of the observation items included in the time-series data set; an adjustment unit that adjusts at least one of a weight dimension of the model used when the extraction unit extracts the feature value from the time-series data set and a corresponding input value, in accordance with a result of the detection by the detection unit; and a storage unit that stores information corresponding to a content of the adjustment by the adjustment unit into a storage device. According to such a configuration, it is possible to respond accurately even if the input dimension changes, and it is possible to support decision-making of a user appropriately even if the input dimension changes.SELECTED DRAWING: Figure 13

Description

[Technical Field]

[0001] The present invention relates to a learning device, a learning method, and a program. [Background technology]

[0002] There is a known technology that enables the system status to be grasped, such as detecting anomalies, by making time-series data searchable.

[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 the 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 a partial time series data set, which is a collection of multiple time series data, into a feature vector that indicates the characteristics of the partial time series data set. Furthermore, the storage unit 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. Then, the search unit selects at least one first feature vector from the multiple first feature vectors stored in the storage unit that is similar to the second feature vector converted from the 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. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2020 / 049666 Summary of the Invention [Problem to be solved by the invention]

[0005] The technology described in Patent Document 1 is based on the premise that the input dimension does not change between learning and searching. However, the input dimension may change between learning and searching due to the addition of new observation items or the deletion of existing observation items. As a result, it becomes necessary to prepare a new model, and re-learning is required. This has led to the problem that it is difficult to achieve efficient machine learning that can accommodate changes in the input dimension.

[0006] Therefore, one of the objects of the present invention is to provide a learning device, a learning method, and a program that can solve the above-mentioned problems. [Means for solving the problem]

[0007] In order to achieve this object, a learning device according to one embodiment of the present disclosure includes: an extraction unit that extracts features from a time series dataset containing time series data for a predetermined observation item using a machine-learned model; a detection unit that detects a change in an input dimension, which is a change in the number of observation items included in the time-series data set; an adjustment unit that adjusts at least one of a weight dimension of a model used when the extraction unit extracts features from the time-series data set or a corresponding input value according to a result of detection by the detection unit; a storage unit that stores information according to the adjustment content by the adjustment unit in a storage device; have The structure is as follows.

[0008] Furthermore, a learning method according to another aspect of the present disclosure includes: The information processing device Using a machine-learned model, features are extracted from a time series dataset containing time series data for a specified observation item; Detect changes in the input dimension, which is a change in the number of observations included in the time series dataset, adjusting at least one of a weight dimension of a model used when extracting features from a time series dataset or a corresponding input value according to the detection result; Store information according to the adjustment content in a storage device. The structure is as follows.

[0009] Furthermore, a program according to another aspect of the present disclosure includes: In the information processing device, Using a machine-learned model, features are extracted from a time series dataset containing time series data for a specified observation item; Detect changes in the input dimension, which is a change in the number of observations included in the time series dataset, adjusting at least one of a weight dimension of a model used when extracting features from a time series dataset or a corresponding input value according to the detection result; Store information according to the adjustment in a storage device. It is a program for realizing the processing. [Effects of the Invention]

[0010] According to the above-mentioned configurations, it is possible to cope with changes in input dimensions. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 10 is a diagram illustrating an example of a problem that arises when making time-series data searchable. [Figure 2] FIG. 2 is a block diagram illustrating an example of the configuration of a learning device. [Figure 3] FIG. 10 is a diagram showing an example of observation item information. [Figure 4] FIG. 10 is a diagram illustrating an example of processing when increasing input dimensions. [Figure 5] FIG. 10 is a diagram for explaining an example of processing when reducing input dimensions. [Figure 6] 10 is a flowchart illustrating an example of the operation of the learning device. [Figure 7] 10 is a flowchart illustrating an example of the operation of the learning device. [Figure 8] FIG. 10 is a block diagram showing another example configuration of the learning device. [Figure 9] FIG. 10 is a diagram illustrating an application example of a learning device. [Figure 10] FIG. 10 is a diagram for explaining an example of an effect of the present disclosure. [Figure 11] FIG. 10 is a diagram illustrating another application example of the learning device. [Figure 12] FIG. 10 is a diagram illustrating an example of a hardware configuration of a learning device according to a second embodiment of the present disclosure. [Figure 13] FIG. 2 is a block diagram illustrating an example of the configuration of a learning device. [Figure 14] 10 is a flowchart illustrating an example of the operation of the learning device. DETAILED DESCRIPTION OF THE INVENTION

[0012] [First embodiment] A first embodiment of the present invention will be described with reference to FIGS. 1 to 11. FIG. 1 is a diagram illustrating an example of a problem that arises when time-series data is made searchable. FIG. 2 is a block diagram illustrating an example of the configuration of a learning device 100. FIG. 3 shows an example of information included in observation item information 144. FIG. 4 is a diagram illustrating an example of processing when increasing input dimensions. FIG. 5 is a diagram illustrating an example of processing when decreasing input dimensions. FIGS. 6 and 7 are flowcharts illustrating an example of operation of the learning device 100. FIG. 8 is a block diagram illustrating another example of the configuration of a learning device. FIG. 9 is a diagram illustrating an example of application of the learning device. FIG. 10 is a diagram illustrating an example of the effect of the present disclosure. FIG. 11 is a diagram illustrating another example of application of the learning device. Note that in the present disclosure, the drawings may be associated with one or more embodiments.

[0013] In a first embodiment of the present disclosure, a learning device 100 is described, which is an information processing device that learns or updates a machine-learned model based on a time-series data set including one or more pieces of time-series data, which are time-series observation data. For example, the input dimensions may change from the time of model learning due to the addition of new observation items, such as an increase in sensors, or the deletion of existing observation items, such as a decrease in sensors. In such a case, as shown in FIG. 1, if a new model corresponding to the changed number of input dimensions is prepared, each weight included in the model will be randomly initialized, and re-learning will be necessary from the beginning.

[0014] Therefore, when the learning device 100 described in this disclosure detects a change in the input dimension, such as an increase or decrease in the number of sensors, it changes the weight dimension of the first layer (input gate) in response to the detected change. In other words, the learning device 100 expands or contracts the weights in response to the change in the input dimension, and continues to use the common part. This allows the learning device 100 to respond to the change in the input dimension without preparing a new model.

[0015] Note that when reducing the input dimension, the learning device 100 may adjust the input value, such as by setting the corresponding input value to zero, instead of adjusting the weight dimension. For example, as shown in Equation 1, there is no mathematical difference between the value propagated to the intermediate layer when deleting the weight (setting it to 0) and when adjusting the input value to 0 (weight W Ψ and the input value x Ψ (Any of the weights can be set to 0.) Therefore, instead of adjusting the weight dimensions, the input values ​​can be adjusted.

number

[0016] Furthermore, when making adjustments in response to changes in input dimensions, such as adjusting weights or input values ​​in response to a reduction in input dimensions, the learning device 100 stores information indicating the reduced observation items, such as sensors, and information corresponding to the adjustment, such as weight values ​​to be deleted. This allows the learning device 100 to make adjustments using past adjustment results in response to changes in input dimensions, such as making adjustments using past information when a reduced observation item is added again. As a result, the learning device 100 can make more efficient adjustments using past learning results.

[0017] In the present disclosure, a change in input dimension refers to a change in the number of time series data included in a time series dataset due to, for example, adding a new observation item or deleting an existing observation item. In other words, a change in input dimension refers to a change in the number of observation items. For example, the number of time series data included in a time series dataset changes in response to a change in the number of sensors acquiring time series data or a change in the number of data input as time series data. In other words, the input dimension changes. The change in input dimension may be either an increase or a decrease, or may occur simultaneously at the same timing.

[0018] In addition, in the present disclosure, it is assumed that the observation content of common observation items remains unchanged before and after a change in the input dimension. In other words, it is assumed that observation items other than the added or deleted observation item continue to acquire time-series data unchanged from before the addition or deletion of the observation item. By changing the weighting dimension in response to a change in the input dimension, the learning device 100 can take into account the influence of a change in the observation item while maintaining the relationship between common inputs before and after the change.

[0019] Furthermore, the learning device 100 described in this disclosure can train a model using, for example, a supervised distance learning method such as that described in Patent Document 1. For example, the learning device 100 randomly extracts anchor samples from a feature group consisting of multiple feature values ​​extracted from a time-series dataset, and also extracts positive samples that are assigned the same label as the anchor sample and negative samples that are assigned a label different from that of the anchor sample. The learning device 100 then updates weight parameters so as to shorten the distance between the anchor sample and the positive sample and to lengthen the distance between the anchor sample and the negative sample in the feature space, thereby training and updating the model. Note that the learning device 100 may also perform training using methods other than those exemplified above.

[0020] Fig. 2 shows an example configuration of learning device 100. Referring to Fig. 2, learning device 100 has, as main components, for example, an operation input unit 110, a screen display unit 120, a communication interface unit 130, a storage unit 140, and an arithmetic processing unit 150.

[0021] 2 illustrates an example in which the functions of learning device 100 are realized using a single information processing device. However, at least some of the functions of learning device 100 may be realized using multiple information processing devices, for example, on the cloud. Furthermore, learning device 100 may not include some of the components illustrated above, such as not having operation input unit 110 or screen display unit 120, or may have components other than those illustrated above.

[0022] Operation input unit 110 is made up of operation input devices such as a keyboard, a mouse, etc. Operation input unit 110 detects operations by the operator operating learning device 100 and outputs the operations to calculation processing unit 150.

[0023] The screen display unit 120 is composed of a screen display device such as a liquid crystal display, an organic EL (electro-luminescence) display, etc. The screen display unit 120 can display various information stored in the storage unit 140 on the screen in response to instructions from the arithmetic processing unit 150.

[0024] The communication interface unit 130 is composed of a data communication circuit, etc. The communication interface unit 130 performs data communication with various sensors and other external devices connected via communication lines.

[0025] The storage unit 140 is a storage device such as a hard disk or a memory. The storage unit 140 stores processing information and a program 145 required for various processes in the arithmetic processing unit 150. The program 145 is read into the arithmetic processing unit 150 and executed to realize various processing units. The program 145 is read in advance from an external device or a recording medium via a data input / output function such as the communication interface unit 130, and is stored in the storage unit 140. Main information stored in the storage unit 140 includes, for example, model information 141, feature information 142, time-series data information 143, and observation item information 144.

[0026] The model information 141 includes information about a model that extracts and outputs features that preserve local distance relationships for inputs of time series segments obtained by dividing a time series dataset. For example, the model information 141 may include weight parameters included in the trained model described above. For example, the model included in the model information 141 is trained in advance using training time series segments inside or outside the training device 100 and stored in the storage unit 140. The model information 141 is updated through training by a model training unit 154 (described later). The number of dimensions of the weight parameters included in the model information 141 is adjusted by an adjustment unit 156 (described later).

[0027] The feature information 142 includes information corresponding to features extracted from the time-series segments. For example, the feature information 142 includes binary codes obtained by converting features using any method. In the feature information 142, the binary codes may be associated with labels to be classified. For example, the binary codes included in the feature information 142 are obtained by converting the features extracted by the feature extraction unit 152 using the binary code conversion unit 153, and are stored in the storage unit 140.

[0028] The information included in the feature amount information 142 can be used when performing a search process, which will be described later. For example, the results of the search process may be used to determine whether an abnormality exists or to determine the label to which the abnormality belongs. The results of the search process may also be used for purposes other than those exemplified above.

[0029] The time-series data information 143 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. The time-series data may be time-series information input using the operation input unit 110, such as information indicating meal times and contents, body temperature, medication contents, etc. The time-series data information 143 is updated in response to the time-series data acquisition unit 151 (described later) acquiring time-series data from sensors and other devices.

[0030] Observation item information 144 includes information on observation items such as sensors, and information corresponding to the status of the observation items and past learning results, such as weight values ​​and input values ​​adjusted in response to changes in the observation items. Observation item information 144 is updated in response to the storage by observation item storage unit 157 of information indicating the adjustment content in response to adjustment by adjustment unit 156. At least a portion of the information included in observation item information 144 may be acquired in advance by receiving it from an external device via communication interface unit 130, and stored in storage unit 140.

[0031] FIG. 3 shows an example of the observation item information 144. Referring to FIG. 3, the observation item information 144 associates, for example, a name, a physical quantity (unit), a standardized parameter, and a learning weight value. Here, the name is information indicating the object being observed, such as an observation item name such as a sensor name or any other arbitrary identifier. The physical quantity (unit) is information indicating a measurement unit, such as g, kg, or db. The standardized parameter indicates a value before adjustment of an input value that has been adjusted to 0 in accordance with a change in the input dimension. The learning weight value indicates a weight value that has been deleted in accordance with a change in the input dimension. Note that items other than those exemplified above may be associated with the observation item information 144.

[0032] The arithmetic processing unit 150 has an arithmetic device such as a CPU (Central Processing Unit) and its peripheral circuits. The arithmetic processing unit 150 reads and executes a program 145 from the storage unit 140, thereby causing the above hardware and the program 145 to work together to realize various processing functions. Major processing units realized by the arithmetic processing unit 150 include, for example, a time-series data acquisition unit 151, a feature extraction unit 152, a binary code conversion unit 153, a model learning unit 154, an input dimension change detection unit 155, an adjustment unit 156, and an observation item information storage unit 157.

[0033] In addition, the arithmetic processing unit 150 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 thereof, instead of the above-mentioned CPU.

[0034] The time series data acquiring unit 151 acquires a time series data set by acquiring time series data from various sensors and other devices. The time series data acquiring unit 151 may accept input of information using the operation input unit 110 or the like as at least a part of the time series data included in the time series data set. Furthermore, the time series data acquiring unit 151 stores the acquired time series data set in the storage unit 140 as time series data information 143.

[0035] The feature extraction unit 152 extracts features from the time series segments by inputting the time series segments to a model stored as model information 141. For example, the feature extraction unit 152 divides the time series data set acquired by the time series data acquisition unit 151 into multiple time series segments using a time window of a certain period. Then, the feature extraction unit 152 inputs each of the divided time series segments to a trained model to extract features. Note that the feature extraction unit 152 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 152 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.

[0036] The binary code conversion unit 153 converts the features extracted by the feature extraction unit 152 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 153 may convert the features extracted by the feature extraction unit 152 into binary code using any method. Furthermore, the binary code conversion unit 153 can store the converted binary code in the storage unit 140 as feature information 142.

[0037] The feature extraction unit 152 or the binary code conversion unit 153 may assign labels to be classified to the features or binary codes using any method. For example, the feature extraction unit 152 or the binary code conversion unit 153 may assign labels to the features, etc., determined based on the Euclidean distance between the time-series segment from which the features were extracted and the time-series segment to be compared. Information about the labels may be acquired based on an operation by the operator on the operation input unit 110, etc.

[0038] The model learning unit 154 performs learning based on the feature amounts extracted by the feature amount extraction unit 152, thereby updating the model stored as the model information 141.

[0039] For example, the model training unit 154 trains the model using a supervised metric learning method such as that described in Patent Document 1. The model training unit 154 can train the model so as to minimize the metric learning loss. As an example, the model training unit 154 randomly extracts anchor samples from a feature group including multiple feature values ​​extracted by the feature extraction unit 152, and also extracts positive samples that are assigned the same label as the anchor sample and negative samples that are assigned a label different from that of the anchor sample. In this case, the model training unit 154 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. Thereafter, the model training unit 154 updates the model by comparing the extracted samples. For example, the model training unit 154 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.

[0040] The labels corresponding to the respective feature amounts may be assigned by the feature amount extraction unit 152 or the binary code conversion unit 153, or may be assigned by the model learning unit 154. For example, the model learning unit 154 may assign a label to each feature amount using a method similar to that used by the feature amount extraction unit 152 or the binary code conversion unit 153.

[0041] The loss function used by the model learning unit 154 during learning may be any function as long as it preserves the 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 learned by the model learning unit 154 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.

[0042] The input dimension change detection unit 155 detects changes in the input dimension. The input dimension change detection unit 155 may use any means to detect changes in the input dimension in response to addition or deletion of an observation item. Furthermore, the input dimension change detection unit 155 can acquire information about the added or deleted observation item.

[0043] For example, the input dimension change detection unit 155 can detect a change in the input dimension in response to receiving information indicating a change in an observation item, such as an increase or decrease in the number of sensors, from management software that manages the sensors that perform observations. The input dimension change detection unit 155 may also detect a change in the input dimension in response to monitoring the number of time-series data acquired by the time-series data acquisition unit 151. The input dimension change detection unit 155 may also detect a change in the input dimension using a method other than those exemplified above, such as detecting a change in the input dimension in response to acquiring information about a new observation item via the operation input unit 110, etc.

[0044] The adjustment unit 156 performs a predetermined adjustment, such as adjusting the weighting dimension of the first layer of the model to be trained, in accordance with the result of detection by the input dimension change detection unit 155. In other words, the adjustment unit 156 increases or decreases the number of weighting parameters of the first layer included in the model information 141 in accordance with the result of detection by the input dimension change detection unit 155. For example, the adjustment unit 156 can expand the weighting dimension in response to detection of an increase in the input dimension, or reduce the weighting dimension in response to detection of a decrease in the input dimension. Furthermore, the adjustment unit 156 may adjust the standardized input value to 0 instead of reducing the weighting dimension of the first layer in response to detection of a decrease in the input dimension.

[0045] For example, referring to FIG. 4, the adjustment unit 156 refers to the observation item information 144 in response to the detection of an increase in the input dimension by the input dimension change detection unit 155. As a result, the adjustment unit 156 checks whether the added observation item is one that was deleted in the past. Then, if it is determined from the observation item information 144 that a new observation item has been added, the adjustment unit 156 expands the weight dimension of the first layer. For example, the adjustment unit 156 may add a weight parameter in accordance with the number of changes in the input dimension, such as an increase in the number of sensors, to each node in the first layer. Here, the weight parameter W to be added is new The value of is, for example, a randomly initialized value. As a result, the output of each input gate is affected by the increase in input dimension, such as the influence of newly added sensors, and is calculated as Wx+W new x new In the above formula, W corresponds to the weight parameter, and x corresponds to the time series data obtained from the sensor. new corresponds to an additional weight parameter that is randomly initialized, and x new This corresponds to time series data acquired by sensors with increased weight parameters. new The value of may be arbitrarily determined based on domain knowledge or the like.

[0046] Furthermore, when it is determined that an observation item such as an added sensor is an observation item that was deleted in the past according to the observation item information 144, the adjustment unit 156 can make adjustments according to the past information included in the observation item information 144. For example, when the weight dimension has been reduced in response to the detection of a decrease in the input dimension, the adjustment unit 156 refers to the observation item information 144 to identify the deleted weight value. Then, the adjustment unit 156 adjusts the weight parameter W to be added. new The specified weight value can be used as the value of . Furthermore, if the input value has been adjusted in response to the detection of a reduction in the input dimension, the adjustment unit 156 refers to the observation item information 144 to specify the input value before adjustment. Then, the adjustment unit 156 can perform standardization or the like using the specified input value to obtain a new input value.

[0047] 5, in response to the detection of a decrease in the input dimension by the input dimension change detection unit 155, such as a decrease in the number of sensors, the adjustment unit 156 reduces the weight dimension of the first layer. For example, the adjustment unit 156 reduces the number of weight parameters at the node corresponding to the input, such as the decreased sensor. As a result, the output of each input gate is affected by the decrease in the input dimension, such as the effect of the decreased sensor, and the W Ω x Ω =Wx-W (s-Ω) x (s-Ω) The adjustment unit 156 may identify the corresponding nodes by any means, such as by storing each sensor in association with the corresponding node. In other words, the adjustment unit 156 may perform the above identification in accordance with the observation item information 144 or the like. The adjustment unit 156 may adjust the weight parameters for each node in the first layer. In the illustrated formula, s corresponds to the set of sensors before reduction, and Ω corresponds to the set of remaining sensors.

[0048] Furthermore, as described above, instead of reducing the weight dimension of the first layer in response to detection of a decrease in input dimension, the adjustment unit 156 can adjust the standardized input value to 0. Whether the adjustment unit 156 adjusts the weight dimension or the input value may be determined arbitrarily.

[0049] For example, as exemplified above, the adjustment unit 156 adjusts the weighting dimension of the first layer of the model to be trained, etc., in accordance with the result of detection by the input dimension change detection unit 155. Note that the adjustment unit 156 may adjust the weighting dimension in the first layer using a method other than the above example, in accordance with the result of detection by the input dimension change detection unit 155. For example, the adjustment unit 156 may adjust the weighting dimension using a method in accordance with the type of model to be trained, such as LSTM.

[0050] The observation item information storage unit 157 stores information indicating the adjustment content and the like in the storage unit 140 as observation item information 144 in response to the adjustment made by the adjustment unit 156. For example, when the adjustment unit 156 makes an adjustment in response to detection of a decrease in input dimensions, the observation item information storage unit 157 stores information corresponding to the adjustment content in the storage unit 140 as observation item information 144. As an example, the observation item information storage unit 157 stores each piece of information included in the observation item information 144, such as information about the observation item corresponding to the adjustment, and the weight value and input value before being adjusted to 0, in the storage unit 140 as information corresponding to the adjustment content. Note that the observation item information storage unit 157 may store information corresponding to the adjustment content made by the adjustment unit 156 in the storage unit 140 as observation item information 144 at timing other than the above-mentioned examples, such as when the adjustment unit 156 makes an adjustment in response to detection of an increase in input dimensions.

[0051] The above is an example of the configuration of learning device 100. Next, an example of the operation of learning device 100 when the input dimension changes will be described with reference to Figs.

[0052] Fig. 5 shows an example of the operation of the learning device 100. Referring to Fig. 5, the input dimension change detection unit 155 detects a change in the input dimension (step S101). The input dimension change detection unit 155 may use any means to detect a change in the input dimension in response to addition or deletion of an observation item, etc.

[0053] The adjustment unit 156 performs adjustment according to the change in dimension in response to the detection by the input dimension change detection unit 155 (step S102). For example, the adjustment unit 156 can adjust at least one of the weight or the input value in response to the detection by the input dimension change detection unit 155.

[0054] In response to the adjustment made by the adjustment unit 156, the observation item information storage unit 157 stores information indicating the adjustment content and the like as the observation item information 144 in the storage unit 140 (step S103).

[0055] The above is an example of the operation of learning device 100. Next, the processing of step S102 will be described in more detail with reference to FIG.

[0056] 7, in response to detection of a decrease in input dimensions by the input dimension change detection unit 155, such as a decrease in the number of sensors (step S201, decrease), the adjustment unit 156 reduces the weight dimension of the first layer or adjusts the standardized input value to 0 (step S202). It may be determined arbitrarily whether the adjustment unit 156 adjusts the weight or the input value.

[0057] In addition, in response to detection of an increase in input dimensions by the input dimension change detection unit 155, such as an increase in sensors (step S201, increase), the adjustment unit 156 refers to the observation item information 144 to check whether the added observation item was previously deleted (step S203).

[0058] If it is determined that the observation item has been deleted from the observation item information 144 in the past (step S203, Yes), the adjustment unit 156 performs adjustment according to the past information included in the observation item information 144 (step S204). On the other hand, if a new observation item is being added (step S203, No), the adjustment unit 156 extends the weight dimension of the first layer by adding a randomly initialized weight parameter or the like (step S205).

[0059] The above is a more detailed example of the operation of step S102. Note that Fig. 7 shows an example of the processing of step S102, and the operation of adjustment unit 156 in step S102 is not limited to the example shown in Fig. 7. For example, the change in the input dimension does not have to be either an increase or a decrease, but may be an increase and a decrease simultaneously. In this case, adjustment unit 156 can simultaneously perform adjustment according to an increase in the input dimension and adjustment according to a decrease in the input dimension.

[0060] As described above, learning device 100 includes input dimension change detection unit 155 and adjustment unit 156. With this configuration, adjustment unit 156 can adjust the weighting dimension of the first layer of the model to be learned, depending on the detection result by input dimension change detection unit 155. As a result, learning device 100 can respond to changes in the input dimension while continuing to use the common part by making adjustments in accordance with changes in the input dimension.

[0061] Furthermore, by referring to observation item information 144, learning device 100 can make adjustments using past information when it is determined that a previously deleted observation item has been restored, etc. As a result, learning device 100 can make more efficient adjustments using past learning results.

[0062] The configuration of the learning device 100 is not limited to the configuration exemplified with reference to Fig. 2. For example, the learning device 100 may be configured with a portion of the configuration exemplified in Fig. 2, such as omitting the binary code conversion unit 153. Fig. 8 shows another example configuration of the learning device 100. Referring to Fig. 8, for example, the arithmetic processing unit 150 can implement a search unit 158 ​​and an output unit 159 in addition to the configuration exemplified in Fig. 2 by reading and executing the program 145 from the storage unit 140.

[0063] The search unit 158 ​​performs search processing based on the feature amount extracted from the time-series segment to be searched.

[0064] For example, when the time-series data acquisition unit 151 acquires a time-series data set to be searched, the feature extraction unit 152 divides the time-series data set to be searched into multiple time-series segments and extracts features. Furthermore, the binary code conversion unit 153 converts the features extracted by the feature extraction unit 152 into binary codes, which are information corresponding to the features. The search unit 158 ​​acquires the binary codes converted as described above. Then, the search unit 158 ​​searches the feature information 142 for binary codes similar to the acquired binary code. For example, the search unit 158 ​​calculates the distance between the acquired binary code and each binary code included in the feature information 142. Then, the search unit 158 ​​acquires 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 142, as a binary code similar to the acquired binary code.

[0065] The search unit 158 ​​can also perform various processes according to the search results. For example, the search unit 158 ​​can infer the label of the time-series segment to be searched by identifying a label associated with the binary code acquired by the search. The search unit 158 ​​can also detect an anomaly according to the search results. For example, a binary code corresponding to time-series data at the time of an anomaly is stored in advance in the feature information 142. The search unit 158 ​​calculates an anomaly score according to the distance between the binary code to be searched and the searched binary code. The search unit 158 ​​can then detect an anomaly based on the calculated anomaly score. For example, the search unit 158 ​​can detect an anomaly based on a comparison result between the calculated anomaly score and a predetermined threshold. The search unit 158 ​​can also perform processes according to search results other than those exemplified above.

[0066] The output unit 159 outputs the search results obtained by the search unit 158. For example, the output unit 159 may output the estimated label, anomaly score, or the like. The output unit 159 can display the search results on the screen display unit 120, or transmit the search results to an external device via the communication interface unit 130. Note that the output unit 159 may be configured to output information stored in the storage unit 140, such as the feature amount information 142, in addition to the search results.

[0067] Furthermore, the learning device 100 described in this disclosure can be applied to various situations in which the input dimension may change. For example, the learning device 100 can be applied to the field of optical communications, such as when estimating the OSNR (Optical Signal-to-Noise Ratio) based on the intensity of an optical signal acquired by an optical transponder or when measuring the degree of anomaly based on the intensity of an optical signal acquired by multiple optical performance monitors in an optical network 200 shown in FIG. 9 . For example, a large-scale network such as the optical network 200 has multiple paths, but the number of paths in the optical network 200 may fluctuate during operation. In this case, the number of optical performance monitors, which are the number of sensors, also fluctuates in conjunction with the number of paths, and the input dimension changes in response to the change in the number of paths. The learning device 100 described in this disclosure can be used to smoothly adapt to changes in the input dimension during use, such as those exemplified above.

[0068] FIG. 10 shows an example of the effect of the learning device 100 when new learning is performed for a predetermined period after the number of optical performance monitors that measure the intensity of optical signals is increased (i.e., when the input dimension is increased). For example, FIG. 10 shows a comparison example between a case in which a new model is prepared in response to a change in the input dimension, each weight parameter is randomly initialized, and learning is then performed, and a case in which continuous learning is performed using the learning device 100. Here, in FIG. 10, the precision is a value such as Precision@1, and is a value that indicates the extent to which the features are captured in binary coding. A larger precision value indicates more accurate and close conversion. Furthermore, the progress of learning is a value such as Triplet loss, and indicates how far learning has progressed. A smaller value indicates more progress in learning. Referring to FIG. 10, it can be seen that the case in which the learning device 100 is used achieves more accurate and close conversion than the case in which a new model is prepared, and that learning has also progressed.

[0069] Furthermore, as shown in FIG. 11 , the learning device 100 may be used in healthcare settings. For example, the learning device 100 can acquire time-series data such as biometric data using one or more biometric sensors 300 possessed by a patient or other user, or other inputs using the operation input unit 110. In such a configuration, the input dimension may change during use, for example, by adding a biometric sensor 300 such as a blood pressure monitor or heart rate monitor, which may increase the types of acquired biometric data, such as heart rate and blood pressure. Furthermore, the types of acquired biometric data may increase or decrease during sleep, work, eating, exercise, etc. The learning device 100 described in this disclosure can easily adapt to changes in input dimension during use, such as those exemplified above. In other words, the learning device 100 described in this disclosure can easily detect anomalies even when the input dimension changes during use, thereby supporting the decision-making of users such as patients. In addition, by realizing a function for sharing detection results with others, such as by transmitting the results to an external device, it is possible to provide better support for decision-making, such as enabling appropriate intervention by family members or doctors in charge.

[0070] As described above, the learning device 100 can be applied to various situations. For example, the learning device 100 may be applied to situations other than the above-described examples, such as a plant.

[0071] [Second embodiment] Next, a second embodiment of the present disclosure will be described with reference to Fig. 12 to Fig. 14. Fig. 12 is a diagram illustrating an example of the hardware configuration of a learning device 400. Fig. 13 is a block diagram illustrating an example of the configuration of the learning device 400. Fig. 14 is a flowchart illustrating an example of the operation of the learning device 400.

[0072] In the second embodiment of the present disclosure, a learning device 400 that is an information processing device that performs machine learning using a time-series dataset will be described. Fig. 12 shows an example of the hardware configuration of the learning device 400. Referring to Fig. 12, the learning device 400 has, as an example, the following hardware configuration. ·CPU (Central Processing Unit) 401 (computing unit) ROM (Read Only Memory) 402 (storage device) RAM (Random Access Memory) 403 (storage device) Programs 404 loaded into RAM 403 A storage device 405 for storing the program group 404 A drive device 406 that reads and writes data from a recording medium 410 outside the information processing device A communication interface 407 for connecting to a communication network 411 outside the information processing device Input / output interface 408 for inputting and outputting data Bus 409 connecting each component

[0073] 13. The learning device 400 can realize the functions of an extraction unit 421, a detection unit 422, an adjustment unit 423, and a storage unit 424 shown in FIG. 13 by having the CPU 401 acquire and execute the program group 404. The program group 404 is stored in advance in, for example, the storage device 405 or the ROM 402, and is loaded into the RAM 403 or the like by the CPU 401 for execution as needed. The program group 404 may be supplied to the CPU 401 via the communication network 411, or may be stored in advance in the recording medium 410, and the drive device 406 may read out the programs and supply them to the CPU 401.

[0074] 12 shows an example of the hardware configuration of the learning device 400. The hardware configuration of the learning device 400 is not limited to the above. For example, the learning device 400 may be configured with only a part of the above configuration, such as excluding the drive device 406. Furthermore, the CPU 401 may be a GPU, as exemplified in the first embodiment.

[0075] The extraction unit 421 uses a machine-learned model to extract features from a time-series data set that includes time-series data for a predetermined observation item.

[0076] The detection unit 422 detects a change in the input dimension, which is a change in the number of observation items included in the time-series dataset. For example, the detection unit 422 may detect a change in the input dimension according to the number of time-series data included in the time-series dataset.

[0077] The adjustment unit 423 adjusts at least one of the weight dimension or the corresponding input value of the model used when the extraction unit 421 extracts features from the time-series data set, according to the detection result by the detection unit 422. Furthermore, the adjustment unit 423 may perform the adjustment using information indicating past adjustment contents stored in the storage unit 424.

[0078] The storage unit 424 stores in a storage device information corresponding to the adjustment content made by the adjustment unit 423. For example, the storage unit 424 may store in a storage device information indicating the corresponding observation item and information corresponding to the adjustment content.

[0079] The above is an example of the configuration of the learning device 400. Next, an example of the operation of the learning device 400 when the input dimension changes will be described with reference to FIG.

[0080] 14 is a flowchart showing an example of the operation of the learning device 400. Referring to Fig. 14, the detection unit 422 detects a change in the input dimension, which is a change in the observation item included in the time-series data set (step S301). The detection unit 422 may detect the change in the input dimension using any means.

[0081] The adjustment unit 423 adjusts the weighting dimension of the model according to the result of detection by the detection unit 422 (step S302).

[0082] The storage unit 424 stores information according to the adjustment content by the adjustment unit 423 in a storage device.

[0083] The above is an example of the operation of the learning device 400 when the input dimension changes.

[0084] As described above, the learning device 400 includes the detection unit 422 and the adjustment unit 423. With this configuration, the adjustment unit 423 can perform predetermined adjustments, such as adjusting the weighting dimensions of the model, in accordance with the detection results of the detection unit 422. As a result, the learning device 400 can respond to changes in the input dimensions while continuing to use the common parts by performing adjustments in accordance with changes in the input dimensions.

[0085] Furthermore, by referring to the information stored in storage unit 424, learning device 400 can make adjustments using past information when it is determined that a previously deleted observation item has been restored, etc. As a result, learning device 400 can make more efficient adjustments using past learning results.

[0086] The above-described learning device 400 can be realized by incorporating a predetermined program into an information processing device such as the learning device 400. 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 400 to implement the following processes: extracting features from a time-series dataset including time-series data on predetermined observation items using a machine-learned model; detecting a change in an input dimension, which is a change in the number of observation items included in the time-series dataset; adjusting at least one of the weight dimension of the model used when extracting features from the time-series dataset or the corresponding input value according to the detection result; and storing information according to the adjustment content in a storage device.

[0087] Furthermore, a learning method executed by an information processing device such as the above-described learning device 400 is a method in which the information processing device uses a machine-learned model to extract features from a time-series dataset including time-series data for predetermined observation items, detects a change in the input dimension, which is a change in the number of observation items included in the time-series dataset, adjusts at least one of the weight dimension of the model used when extracting features from the time-series dataset or the corresponding input value according to the detection result, and stores information according to the adjustment content in a storage device.

[0088] Even if the invention is a program having the above-mentioned configuration, or a computer-readable recording medium having the program recorded thereon, or a learning method, it can achieve the same functions and effects as the above-mentioned learning device 400, and therefore can achieve the above-mentioned objective of the present disclosure.

[0089] <Additional Notes> A part or all of the above-described embodiments can be described as follows: The learning device and the like according to the present invention will be outlined below. However, the present invention is not limited to the following configuration.

[0090] (Appendix 1) an extraction unit that extracts features from a time series dataset containing time series data for a predetermined observation item using a machine-learned model; a detection unit that detects a change in an input dimension, which is a change in the number of observation items included in the time-series data set; an adjustment unit that adjusts at least one of a weight dimension of a model used when the extraction unit extracts features from the time-series data set or a corresponding input value according to a result of detection by the detection unit; a storage unit that stores information according to the adjustment content by the adjustment unit in a storage device; have Learning device. (Appendix 2) 10. The learning device according to claim 1, The storage unit stores information corresponding to the adjustment content in the storage device when the adjustment unit performs a predetermined adjustment in response to the detection of a reduction in input dimensions by the detection unit. Learning device. (Appendix 3) 3. The learning device according to claim 2, The adjustment unit checks whether the added observation item is one that has been deleted in the past by referring to the information stored in the storage unit in response to the detection by the detection unit of an increase in input dimensions. Learning device. (Appendix 4) 4. The learning device according to claim 3, When the added observation item is one that has been deleted in the past, the adjustment unit adjusts at least one of the weight dimension of the model or the corresponding input value using the information stored in the storage unit. Learning device. (Appendix 5) 5. The learning device according to claim 3 or 4, The adjustment unit extends the weight dimension of the model in response to the detection by the detection unit of an increase in input dimension if the added observation item is not one that has been deleted in the past. Learning device. (Appendix 6) 10. The learning device according to claim 1, wherein: The adjustment unit reduces a weight dimension of the model or adjusts a corresponding input value to 0 in response to detection of a reduction in input dimension by the detection unit. Learning device. (Appendix 7) 10. The learning device according to claim 1, wherein: The detection unit detects a change in the number of sensors that acquires time-series data as a change in the input dimension. Learning device. (Appendix 8) 10. The learning device according to claim 1, wherein: a learning unit that performs machine learning to update weights of a model using the feature values ​​extracted by the extraction unit; a conversion unit that converts the feature quantity extracted by the extraction unit into a binary code; a storage unit that stores the binary code converted by the conversion unit; a search unit that searches for binary codes stored in the storage unit by using binary codes converted from a time series data set to be searched; have Learning device. (Appendix 9) The information processing device Using a machine-learned model, features are extracted from a time series dataset containing time series data for a specified observation item; Detecting a change in input dimension, which is a change in the number of observation items included in the time series data set; adjusting at least one of a weight dimension of a model used in extracting features from the time-series data set or a corresponding input value according to the detection result; Store information according to the adjustment in a storage device. How to learn. (Appendix 10) In the information processing device, Using a machine-learned model, features are extracted from a time series dataset containing time series data for a specified observation item; Detecting a change in input dimension, which is a change in the number of observation items included in the time series data set; adjusting at least one of a weight dimension of a model used in extracting features from the time-series data set or a corresponding input value according to the detection result; Store information according to the adjustment in a storage device. A program to realize the processing.

[0091] Note that some or all of the configurations described in Supplementary Notes 2 to 8 that are dependent on the learning device described in Supplementary Note 1 may also be dependent in a similar dependent relationship on the learning method described in Supplementary Note 9 and the program described in Supplementary Note 10. Furthermore, not limited to Supplementary Notes 9 and 10, some or all of the configurations described as Supplements may also be dependent on various hardware, software, various recording means for recording software, or systems within the scope of the above-mentioned embodiments.

[0092] 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.

[0093] 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. [Explanation of symbols]

[0094] 100 Learning Device 110 Operation input section 120 Screen display section 130 Communication interface unit 140 Storage section 141 Model Information 142 Feature Information 143 Time Series Data Information 144 Observation Item Information 145 Programs 150 Processing unit 151 Time series data acquisition unit 152 Feature Extraction Unit 153 Binary Code Conversion Unit 154 Model Learning Department 155 Input dimension change detection unit 156 Adjustment section 157 Observation item information storage section 158 Search Department 159 Output section 200 Optical Network 300 Biometric Sensor 400 Learning Device 401 CPU 402 ROM 403 RAM 404 Programs 405 Storage device 406 Drive Unit 407 Communication Interface 408 Input / Output Interface 409 Bus 410 Recording Media 411 Communication Network 421 Extraction part 422 Detection unit 423 Adjustment section 424 Storage Unit

Claims

1. an extraction unit that extracts features from a time series dataset including time series data for a predetermined observation item using a machine-learned model; a detection unit that detects a change in an input dimension, which is a change in the number of observation items included in the time-series data set; an adjustment unit that adjusts at least one of a weight dimension of a model used when the extraction unit extracts features from the time-series data set or a corresponding input value according to a result of detection by the detection unit; a storage unit that stores information according to the adjustment content by the adjustment unit in a storage device; have Learning device.

2. The learning device according to claim 1 , The storage unit stores information corresponding to the adjustment content in the storage device when the adjustment unit performs a predetermined adjustment in response to the detection of a reduction in input dimensions by the detection unit. Learning device.

3. The learning device according to claim 2, The adjustment unit checks whether the added observation item is one that has been deleted in the past by referring to the information stored in the storage unit in response to the detection by the detection unit of an increase in input dimensions. Learning device.

4. The learning device according to claim 3, When the added observation item is a previously deleted item, the adjustment unit adjusts at least one of the weight dimension of the model or the corresponding input value using the information stored in the storage unit. Learning device.

5. The learning device according to claim 3, The adjustment unit extends the weight dimension of the model in response to the detection by the detection unit of an increase in input dimension if the added observation item is not one that has been deleted in the past. Learning device.

6. The learning device according to claim 1 , The adjustment unit reduces a weight dimension of the model or adjusts a corresponding input value to 0 in response to detection of a reduction in input dimension by the detection unit. Learning device.

7. The learning device according to claim 1 , The detection unit detects a change in the number of sensors that acquires time-series data as a change in the input dimension. Learning device.

8. The learning device according to claim 1 , a learning unit that performs machine learning to update weights of a model using the feature values ​​extracted by the extraction unit; a conversion unit that converts the feature quantity extracted by the extraction unit into a binary code; a storage unit that stores the binary code converted by the conversion unit; a search unit that searches for binary codes stored in the storage unit by using binary codes converted from a time series data set to be searched; have Learning device.

9. The information processing device Using a machine-learned model, features are extracted from a time series dataset containing time series data for a specified observation item; Detecting a change in input dimension, which is a change in the number of observation items included in the time series data set; adjusting at least one of a weight dimension of a model used in extracting features from the time-series data set or a corresponding input value according to the detection result; Store information according to the adjustment in a storage device. How to learn.

10. In the information processing device, Using a machine-learned model, features are extracted from a time series dataset containing time series data for a specified observation item; Detecting a change in input dimension, which is a change in the number of observation items included in the time series data set; adjusting at least one of a weight dimension of a model used in extracting features from the time-series data set or a corresponding input value according to the detection result; Store information according to the adjustment in a storage device. A program to realize the processing.

Images