Information processing systems, information processing methods, and programs

Abstract

To provide an information processing system that can easily perform various pre-processing steps. [Solution] According to one aspect of the present invention, an information processing system is provided comprising one or more processors, wherein in a preprocessing step, the processor preprocesses data using a transformation function, the transformation function is a function having adjustable parameters and representing a transformation expression of a type corresponding to the adjusted parameters from among two or more types of transformation expressions, and in an output step, the processor processes the data preprocessed with the transformation function using machine learning to obtain an output.

Description

【Technical Field】 , , 【0004】 , 【0006】 , , , 【0005】 , , , , 【0001】 The present invention relates to an information processing system, an information processing method, and a program. 【Background Art】 【0002】 In Patent Document 1, in a drawing search device, a learning unit generates teacher data including content parameters of each of a plurality of search target drawings by analyzing each of the plurality of search target drawings, the learning unit generates a learning model based on the teacher data, and a search unit uses the learning model to collate a target drawing with each of the plurality of search target drawings, thereby searching for at least one predetermined drawing corresponding to the target drawing from the plurality of search target drawings. A technique is disclosed. 【Prior Art Documents】 【Patent Documents】 【0003】 【Patent Document 1】 Japanese Unexamined Patent Application Publication No. 2021-12413 【Summary of the Invention】 【Problems to be Solved by the Invention】 【0004】 In machine learning, for example, as preprocessing for improving the efficiency and accuracy of learning, a process of converting learning data is performed. There are various types of such conversion processes, and in order to find a conversion type suitable for the learning data, it takes time to change the conversion type used for preprocessing. 【0005】 In view of the above circumstances, the present invention aims to provide an information processing system and the like that can easily perform various preprocessings. [[ID=*41]]【Means for Solving the Problems】 【0006】 It should be noted that in the translation, the 7-digit tags to are preserved as they are, and the text within the tags is not translated as it seems to be some specific identifiers or codes without clear semantic meaning for translation. Also, the Japanese term "発明が解決しようとする課題" is translated as "Problems to be Solved by the Invention", and "課題を解決するための手段" is translated as "Means for Solving the Problems" which are common translations in patent text translations. The term "機械学習" is translated as "machine learning".According to one aspect of the present invention, an information processing system is provided comprising one or more processors, wherein in a preprocessing step, the processor preprocesses data using a transformation function, the transformation function is a function having adjustable parameters and representing a transformation expression of a type corresponding to the adjusted parameters from among two or more types of transformation expressions, and in an output step, the processor processes the data preprocessed with the transformation function using machine learning to obtain an output. 【0007】 This configuration allows for easy execution of various pretreatment processes. [Brief explanation of the drawing] 【0008】 [Figure 1] This is a diagram illustrating an example of the overall configuration of Information Processing System 1. [Figure 2] This is a block diagram showing an example of the hardware configuration of the information processing device 2. [Figure 3] This block diagram shows an example of the hardware configuration of user terminal 3. [Figure 4] This is a flowchart illustrating an example of the learning process flow. [Figure 5] This figure shows an example of evaluating a learning model. [Figure 6] This is a flowchart illustrating an example of the task execution process flow. [Figure 7] This figure shows an example of recorded parameters. [Modes for carrying out the invention] 【0009】 Embodiments of the present invention will be described below with reference to the drawings. The various features shown in the embodiments below can be combined with each other. 【0010】 Incidentally, the program for implementing the software appearing in one embodiment may be provided as a non-transitory computer-readable medium, or it may be provided as a downloadable medium from an external server, or it may be provided so that the program is launched on an external computer and its functions are realized on a client terminal (so-called cloud computing). 【0011】 Furthermore, in various information processing according to one embodiment, an input and an output corresponding to the input can be realized. Here, as long as an output is obtained as a result of the input, the form of the information referenced in such information processing (hereinafter referred to as "reference information") is not limited. The reference information may be, for example, rule-based information such as a database, a lookup table, or a predetermined function (including a decision formula such as a regression equation constructed by a statistical method), or a pre-trained model that has learned the correlation between input and output in advance, or a large-scale language model that can output a desired result by inputting a prompt. 【0012】 Furthermore, in one embodiment, "part" may include, for example, hardware resources implemented by a circuit in a broad sense, and the information processing of software that can be specifically realized by these hardware resources. Also, in one embodiment, various types of information are handled, and this information can be represented, for example, by the physical values ​​of signal values ​​representing voltage and current, the high or low values ​​of signal values ​​as a set of binary bits composed of 0s or 1s, or by quantum superposition (so-called qubits), and communication and calculations can be performed on a circuit in a broad sense. 【0013】 Furthermore, a circuit in a broad sense is a circuit realized by combining at least a suitable combination of circuits, circuits, processors, and memory. The processor may be a general-purpose processor or a dedicated circuit. In other words, it includes application-specific integrated circuits (ASICs), programmable logic devices (for example, simple programmable logic devices (SPLDs), complex programmable logic devices (CPLDs), and field programmable gate arrays (FPGAs)), etc. 【0014】 1. Hardware Configuration This section describes the hardware configuration. 【0015】 <Information Processing System 1> Figure 1 is a diagram showing an example of the overall configuration of information processing system 1. Information processing system 1 comprises an information processing device 2 and a user terminal 3. The information processing device 2 and the user terminal 3 are configured to communicate with each other via a telecommunications line. In one embodiment, information processing system 1 consists of one or more devices or components. For example, if it consists only of an information processing device 2, then information processing system 1 can be information processing device 2. In the following explanation, a so-called classical computer is used as the information processing device 2, but a quantum computer or a hybrid computer of a classical computer and a quantum computer may also be used. These components will be described below. 【0016】 <Information Processing Device 2> FIG. 2 is a block diagram showing an example of the hardware configuration of the information processing apparatus 2. The information processing apparatus 2 includes a communication unit 21, a storage unit 22, and a processor 23, and these components are electrically connected via a communication bus 20 inside the information processing apparatus 2. Each component will be further described. 【0017】 The communication unit 21 preferably uses wired communication means such as USB, IEEE1394, Thunderbolt (registered trademark), and wired LAN network communication. However, wireless LAN network communication, mobile communication such as 3G / LTE / 5G, and BLUETOOTH (registered trademark) communication may be included as needed. That is, it is more preferable to implement it as a collection of these multiple communication means. That is, the information processing apparatus 2 may communicate various information from the outside via the communication unit 21 and the network. 【0018】 The storage unit 22 stores various information defined as described above. This can be implemented as a storage device such as a solid state drive (SSD) that stores various programs related to the information processing apparatus 2 executed by the processor 23, or as a memory such as a random access memory (RAM) that stores temporarily necessary information (arguments, arrays, etc.) related to program operations. The storage unit 22 stores various programs, variables, etc. related to the information processing apparatus 2 executed by the processor 23. 【0019】 The processor 23 performs the processing and control of the overall operation related to the information processing apparatus 2. The processor 23 is, for example, a central processing unit (CPU) not shown in the figure. The processor 23 realizes various functions related to the information processing apparatus 2 by reading a predetermined program stored in the storage unit 22. That is, the information processing by software stored in the storage unit 22 is specifically realized by the processor 23, which is an example of hardware, and can be executed as each functional unit included in the processor 23. These will be described in more detail in the next section. Note that the processor 23 is not limited to being single, and may be implemented to have a plurality of processors 23 for each function. Or a combination thereof may also be used. 【0020】 <User terminal 3> FIG. 3 is a block diagram showing an example of the hardware configuration of the user terminal 3. The user terminal 3 includes a communication unit 31, a storage unit 32, a processor 33, a display unit 34, and an input unit 35, and these components are electrically connected via a communication bus 30 inside the user terminal 3. The descriptions of the communication unit 31, the storage unit 32, and the processor 33 are omitted because they are the same as the descriptions of the respective units in the information processing apparatus 2. 【0021】 The display unit 34 may be included in the housing of the user terminal 3 or may be externally attached. The display unit 34 displays a screen of a graphical user interface (GUI) operable by the user. For example, it is preferable to selectively use display devices such as a CRT display, a liquid crystal display, an organic EL display, and a plasma display according to the type of the user terminal 3. 【0022】 The input unit 35 may be included in the casing of the user terminal 3 or it may be an external component. For example, the input unit 35 may be integrated with the display unit 34 and implemented as a touch panel. If it is a touch panel, the user can input tap operations, swipe operations, etc. Of course, instead of a touch panel, a switch button, mouse, QWERTY keyboard, etc., may be used. In other words, the input unit 35 receives operation input made by the user. This input is transferred as a command signal to the processor 33 via the communication bus 30, and the processor 33 can execute predetermined controls and calculations as needed. 【0023】 2. Regarding information processing The following describes the information processing according to the embodiment, specifically the learning process, which involves using training data and machine learning algorithms to cause the learning model of the AI ​​(Artificial Intelligence) module to perform machine learning, and the task execution process, in which the AI ​​module executes tasks using the learning model adjusted through the learning process. In the following description, the information processing device 2 and the user terminal 3 are described as the main components of the learning process and the task execution process, but these information processing operations are performed by one or more processors in the control unit of each device. 【0024】 Figure 4 is a flowchart illustrating an example of the learning process flow. Figure 4 shows the flow of processing performed by the information processing device 2. Note that this information processing may include arbitrary exception handling not shown. Exception handling includes interrupting the information processing or omitting individual processes. The selections or inputs made in this information processing may be based on user operation or may be performed automatically without user operation. 【0025】 First, the information processing device 2 acquires training data (step S1). Training data is data used to enable machine learning in the learning model, and includes training data used for parameter optimization, validation data to verify the performance of the learning model, and test data for final performance evaluation. Training data may be multimodal data containing multiple types of data, but in the following explanation, for the sake of clarity, we will describe an example in which single-modal data containing one type of data (image data, text data, and sound data, etc.) is used as training data. 【0026】 The information processing device 2 uses a parameterized generalized transformation function in the learning process. A parameterized generalized transformation function is a generalized function that allows for dynamic application of transformations by setting values ​​(parameters) to control the transformation. For example, the information processing device 2 uses the output X[k] of the generalized transformation function represented by the following equation 1 (where "equation" indicates "mathematical expression"). In equation 1, x[n] represents a discrete-time signal, N represents the number of samples in the signal, and k represents the frequency index. 【0027】 【number】 【0028】 Number 1 contains three factors: the first factor represented by Number 2, the second factor represented by Number 3, and the third factor represented by Number 4. The "I" in Numbers 3 and 4 represents the identity matrix. 【0029】 【number】 【0030】 【number】 【0031】 【number】 【0032】 Equation 3 uses the parameter "α", and Equation 4 uses the parameter "β". When α=0 and β=0, both Equation 3 and Equation 4 become identity matrices, and the generalized transformation function becomes the function represented by Equation 2. This Equation 2 represents the Discrete Fourier Transform (DFT). In other words, X[k] in this case is the output transformed by the DFT represented by the generalized transformation function to which the parameters α=0 and β=0 are applied. 【0033】 Furthermore, if α=1 and β=0, then equation 4 becomes the identity matrix, and the function is represented by equation 5 below. 【0034】 【number】 【0035】 This number 5 represents the Discrete Hartley Transform (DHT). In other words, X[k] in this case is the output transformed by the DHT represented by a generalized transformation function to which the parameters α=1 and β=0 are applied. 【0036】 Furthermore, when α=0 and β=1, equation 3 represents the identity matrix, and the function is represented by equation 6 below. 【0037】 【number】 【0038】 This number 5 represents the Discrete Cosine Transform (DCT-type II). In other words, X[k] in this case is the output transformed by the DCT-type II, which is represented by a generalized transformation function to which the parameters α=0 and β=1 are applied. 【0039】 Next, the user selects several appropriate transformations and defines them in the form of equation 1 (step S2). For example, if the three transformations DFT, DHT, and DCT-typeII are deemed appropriate, then f[n,k] and g[n,k] in equation 1 become equations 3 and 4, respectively. If other transformations are deemed appropriate, then the equations for f[n,k] and g[n,k] should be adjusted. For example, if DFT, DHT, and the Discrete Sine Transform (DST-typeII) are deemed appropriate instead of DFT, DHT, and DCT-typeII, then f[n,k] remains equation 3, and g[n,k] should be changed to equation 7 as follows. 【0040】 【number】 【0041】 If we let g[n,k] be equation 7 and apply the parameters α=0 and β=1, then equation 3 represents the identity matrix, and the generalized transformation function of equation 1 becomes the function shown in equation 8 below. 【0042】 【number】 【0043】 This number 8 represents the Discrete Sine Transform (DST-type II). In other words, when g[n,k] is represented by number 7 and parameters α=0 and β=1 are applied, the output X[k] is the output transformed by DST-type II. As an example, we have defined three transformations: DFT, DHT, and DCT-type II, but the number of transformations can be increased if appropriate. For example, if we want to introduce DST-type II in addition to DFT, DHT, and DCT-type II, we can introduce a new item h[n,k] into number 1 as shown in number 9 below. 【0044】 【number】 【0045】 f[n,k] and g[n,k] are numbers 3 and 4, respectively, and h[n,k] is the following number 10. 【0046】 【number】 【0047】 In other words, equation 1 can freely define two or more types of transformations. In this invention, the operation performed between functions f[n,k] and g[n,k] is not limited to multiplication, but addition can also be applied. In other words, it can also be expressed as the following equation 11 as a general transformation function. 【0048】 【number】 【0049】 In the following, for the sake of simplicity, we consider an example in which three transformations—DFT, DHT, and DCT-type II—are introduced using a generalized transformation function based on Equation 1. Next, random values ​​are assigned to the parameters of the generalized transformation function, and the information processing device 2 performs preprocessing of the training data using the above generalized transformation function (step S3). Preprocessing here may include not only feature extraction but also processes to simplify feature extraction (denoising, standardization, or imputation of missing values, etc.). 【0050】 Next, the information processing device 2 performs machine learning processing using the pre-processed training data and machine learning algorithm using a general transformation function (step S4). For example, the information processing device 2 selects a training model to be used for machine learning, and instructs the machine learning algorithm to input the pre-processed training data into the selected training model and perform machine learning, thereby causing the training model to perform machine learning. The machine learning performed here may be supervised learning or unsupervised learning. 【0051】 Next, the information processing device 2 evaluates the learning model generated by machine learning (step S5). Evaluating the learning model means measuring the accuracy, effectiveness, or efficiency of the machine learning model when performing a task. For example, if supervised learning is performed, the information processing device 2 uses the generated learning model to make predictions, measures the loss, and evaluates the generated learning model more highly the smaller the measured loss value. 【0052】 Figure 5 shows an example of evaluating a learning model. In Figure 5, the horizontal axis represents the number of epochs, and the vertical axis shows a line graph representing the cross-entropy loss, which is an example of loss. In the example in Figure 5, the relationship between the number of epochs and the cross-entropy loss for each learning model, "FNet" (DFTNetwork), "DCTNet" (DCTNetwork), "HartleyNet" (HartleyNetwork), and "GTNet" (General Transform Network), is shown. 【0053】 These learning models represent, respectively, a learning model that uses DFT-transformed training data, a learning model that uses DCT-transformed training data, a learning model that uses Hartley-transformed training data, and a learning model that uses training data transformed using the general transformation function described above. In all of these learning models, the cross-entropy loss decreases as the number of epochs increases, and convergence is generally observed around 50 epochs. 【0054】 The information processing device 2 determines whether the loss calculation has converged (step S6). If it determines that it has not converged (NO), it updates the parameters of the general transformation function and the machine learning parameters (model parameters) (step S7), and returns to step S3 to execute the process again from the preprocessing stage. If the information processing device 2 determines in step S6 that the loss calculation has converged (YES), it determines the optimal parameters (step S8). 【0055】 Here, the example above described the case where the parameters (α,β) are adjusted to (0,0), (1,0), and (0,1). However, they are not limited to these; they may also be adjusted to (1,1), or to decimal numbers such as (0.5,0) or (0.3,0.5). Below, for example, we will assume that the parameters are adjusted to result in 121 combinations when both α and β take on 11 possible values: 0, 0.1, 0.2, ..., 0.9, and 1. Note that the method of adjusting the parameters is not limited to this; they may be adjusted to result in fewer combinations or more combinations. 【0056】 The information processing device 2 determines, for example, the parameters applied to the general transformation function used in the preprocessing of the learning model that minimized the loss loss as the optimal parameters. Then, the information processing device 2 stores the determined optimal parameters (step S9). 【0057】 The accuracy of predictions at convergence was 91.3% for FNet, 95.2% for DCTNet, 96.71% for HartleyNet, and 96.82% for GTNet. The GTNet value shown is the value obtained by adjusting the parameter that yielded the highest evaluation among the parameters adjusted as described above. From these results, it was shown that the training data (testing dataset) used in the example in Figure 5 generated a trained model that performed better when DFT was performed than when DCT was performed, and better when the Hartley transform was performed than when DFT was performed. Furthermore, it was shown that when a general transformation function was used for the transformation (when GTNet was used), it was possible to generate a trained model that performed better than the Hartley transform. 【0058】 This concludes the explanation of the learning process. Next, we will explain the task execution process. The task is executed by an AI module equipped with a learning model. The tasks that the AI ​​module can perform are diverse and include, for example, prediction, analysis, natural language processing, image processing, video processing, control processing, or recommendation processing. The learning model described above has its parameters (model parameters, etc.) adjusted through machine learning performed in the machine learning process to be optimal for performing each task. 【0059】 Figure 6 is a flowchart illustrating an example of the task execution process flow. Figure 6 shows the flow of processing performed by the information processing device 2. Note that this information processing may include any exception handling not shown. Exception handling includes interrupting the information processing or omitting individual processes. The selections or inputs made in this information processing may be based on user operation or may be performed automatically without user operation. 【0060】 First, the information processing device 2 acquires task data (step S11). Task data is data used to execute a task using a trained model. In the example above, since machine learning is performed with single-modal training data, similarly, single-modal data such as image data, text data, or sound data is used for the task data. However, if machine learning is performed with multimodal data, multimodal task data may be used. 【0061】 The information processing device 2 reads the optimal parameters of the general transformation function recorded in step S9 shown in Figure 4 (step S12), and performs preprocessing on the task data using the general transformation function to which the read parameters have been applied (step S13). Subsequently, the information processing device 2 reads the parameters of the learning model that has undergone machine learning processing using the training data that has been preprocessed with the same general transformation function to which the optimal parameters have been applied, i.e., the optimized parameters of the learning model (model parameters) (step S14). 【0062】 The information processing device 2 then inputs task data into an AI module equipped with a learning model to which the loaded optimized parameters have been applied and processes it (step S15), and obtains the output of the AI ​​module as the task execution result (step S16). The output execution result differs depending on the content of the task; for example, if it is a prediction process, it will be a prediction result, and if it is an analysis process, it will be an analysis result. This concludes the explanation of the task execution process. 【0063】 As described above, the information processing device 2 is an example of an information processing system equipped with one or more processors. Furthermore, the information processing device 2 functions as an example of a preprocessing unit that performs data preprocessing using a conversion function. Step S3 shown in Figure 4 and step S13 shown in Figure 6 are examples of preprocessing steps, and the general conversion function described above is an example of a conversion function used for preprocessing. 【0064】 Furthermore, a conversion function is a function that represents a conversion formula. More specifically, a conversion function is a function that has adjustable parameters and represents a type of conversion formula corresponding to the adjusted parameters from among two or more types of conversion formulas. The general conversion function represented by Equations 1, 3, and 4 is an example of a conversion function, and the parameter α represented by Equation 3 and the parameter β represented by Equation 4 are examples of adjustable parameters, respectively. 【0065】 Furthermore, the information processing device 2 functions as an example of an output unit that processes data preprocessed by a transformation function using machine learning to obtain an output. Step S4 shown in Figure 4, and steps S15 and S16 shown in Figure 6 are examples of output steps. In the machine learning process shown in Figure 4, the information processing device 2 performs preprocessing and machine learning processing for each combination of parameters α and β in order to find the optimal parameters. On the other hand, in the task execution process shown in Figure 6, the information processing device 2 applies the recorded optimal parameters to a general transformation function and performs preprocessing and processing by the learned model. 【0066】 For example, when giving instructions to perform machine learning after preprocessing using multiple transformation formulas, it is necessary to create those multiple transformation formulas and create separate machine learning instructions, including preprocessing, for each transformation formula used. In contrast to that case, with the above embodiment, if instructions including preprocessing using a general transformation function are created, then all that is needed is to sequentially change the combination of parameters, making it easy to perform a variety of preprocessing steps. 【0067】 Furthermore, the above generalized transformation functions can result in transformations that are not only conventionally defined (such as DFT, DHT, DCT-type II, and DST-type II), but also, depending on the parameters applied, transformations that have not yet been defined. For example, in the generalized transformation functions represented by Equations 1, 3, and 4, applying the parameters α=0.5 and β=0.5, α=0.7 and β=0.3, or α=0.3 and β=0.7 respectively yields transformations that combine the characteristics of DFT, DHT, and DCT-type II in different proportions. Thus, using the above generalized transformation functions makes it easier to express newly defined transformations compared to not using them. 【0068】 Furthermore, the above transformation function has multiple factors corresponding to different types of transformation formulas. For example, the general transformation function represented by equation 1 above has a first factor represented by equation 2, a second factor represented by equation 3, and a third factor represented by equation 4. The first factor corresponds to DFT, the second factor corresponds to DHT, and the third factor corresponds to DCT-type II. The above transformation function allows for adjustment of the contribution of the transformation formula to the preprocessing by changing the weights of each of the multiple factors using parameters. 【0069】 For example, in the case of the second factor, it has term 1, represented by the following number 12, and term 2, represented by the number 13. Both term 1 and term 2 are included in the aforementioned number 3. 【0070】 【number】 【0071】 【number】 【0072】 Term 1 is the term that becomes the identity matrix when α is 0 (an example of the first value). Term 2 is the term that becomes the DHT corresponding to the second factor, as shown in the above-mentioned number 5, when α is 1 (an example of the second value). In the case of the third factor, it has term 3, represented by the following number 14, and term 4, represented by number 15. Both term 3 and term 4 are included in the above-mentioned number 4. 【0073】 【number】 【0074】 【number】 【0075】 Term 3 is the term that becomes the identity matrix when β is 0 (an example of the first value). Term 4 is the term that becomes DCT-type II, corresponding to the third factor, as shown in number 6 above, when β is 1 (an example of the second value). 【0076】 For example, in the generalized transformation function represented by equation 1, when α=0 and β=0, both the second and third factors become the identity matrix, resulting in a DFT corresponding to the first factor, and the contributions to the preprocessing of DFT, DHT, and DCT-typeII are 100%, 0%, and 0%. Also, in the generalized transformation function represented by equation 1, when α=1 and β=0, the third factor becomes the identity matrix, but the second term of the second factor remains, resulting in a DHT corresponding to the second factor, and the contributions to the preprocessing of DFT, DHT, and DCT-typeII are 0%, 100%, and 0%. 【0077】 Furthermore, in the generalized transformation function represented by equation 1, when α=0 and β=1, the second factor becomes the identity matrix, but the fourth term of the third factor remains, resulting in DCT-typeII corresponding to the third factor, and the contributions of DFT, DHT, and DCT-typeII to preprocessing become 0%, 0%, and 100%, respectively. Also, if both α and β are 1, or if decimals are used, the contributions of DFT, DHT, and DCT-typeII to preprocessing can be D1%, D2%, and D3% (100%≧D1,D2,D3≧0%, D1+D2+D3=100%). 【0078】 In this configuration, the contribution of each transformation formula can be easily changed simply by changing the parameters, compared to the case where the above general transformation formula is not used. 【0079】 Furthermore, the information processing device 2 functions as an example of a determination unit that determines the optimal parameters. Step S8 shown in Figure 4 is an example of a determination step. Specifically, the information processing device 2 determines the parameters for which the evaluation result of the learning model optimized by machine learning was good. The machine learning referred to here is machine learning performed using data that has been preprocessed with a general transformation function to which the parameters are applied. 【0080】 For example, in the above example, the information processing device 2 applies 121 different parameters to the general transformation function to perform machine learning processing, and determines the parameters that result in the smallest loss loss for the optimized learning model (the optimal parameters mentioned above) as good parameters. Note that the parameters that result in the smallest loss loss are not the only ones that are good; parameters that fall within a predetermined order from the smallest loss loss may also be considered good parameters. 【0081】 In the example above, the learning model was evaluated based on cross-entropy loss, but this is not the only method; the evaluation should be tailored to the type of task the learning model optimizes. For example, for classification tasks, evaluation can be performed using metrics such as Accuracy, Precision, Recall, or F1-score, while for regression tasks, evaluation can be performed using metrics such as AIC (Akaike Information Criterion), BIC (Bayesian Information Criterion), MSE (Mean Squared Error), or RMSE (Root Mean Squared Error). 【0082】 In either case, it is possible to identify the parameters that yield a high-performance learning model. High performance, in this context, means obtaining more desirable results on a given task. For example, if the task involves prediction, classification, or anomaly detection, a learning model can be considered high-performance if it produces highly accurate results in those tasks. 【0083】 Furthermore, the information processing device 2 functions as an example of a recording unit that records parameters that have been determined to be good. Step S9 shown in Figure 4 is an example of a recording step. With this configuration, compared to the case where the aforementioned recording unit does not function, it becomes easier to reuse parameters that yield a high-performance learning model. 【0084】 Furthermore, the information processing device 2 (an example of a preprocessing unit) preprocesses the task data using a conversion function that represents a type of conversion formula corresponding to the recorded good parameters. Step S13 shown in Figure 6 is an example of a preprocessing step. The information processing device 2 also functions as an example of an execution unit that inputs the preprocessed task data into a learning model and executes the task. Steps S14 and S15 shown in Figure 6 are examples of execution steps. 【0085】 The learning model is a learning model in which machine learning is performed on training data that has been preprocessed with a general transformation function that represents a type of transformation formula corresponding to the recorded good parameters. In this configuration, the same preprocessing as that performed on a learning model whose machine learning results were highly evaluated is performed, and the learning model generated by performing that preprocessing is used, so the degree of task achievement can be increased compared to when the above preprocessing and execution units do not function. 【0086】 4. Others The above method of information processing is merely an example and is not limited to it. For example, in the above example, the generalized transformation function represents three types of transformation formulas (DFT, DCT-type II, and DHT), while only two types of parameters (α, β) are used. However, it is also possible to use the same three types of parameters as the types of transformation formulas (e.g., α, β, γ). Furthermore, the generalized transformation function may represent two types of transformation formulas, or four or more types. The more types of transformation formulas the generalized transformation function represents, the higher the probability of finding the optimal formula. Therefore, a greater number of types is desirable for finding the optimal formula. 【0087】 Furthermore, the more types of transformations there are, the greater the number of transformation formulas that can be expressed by the general transformation function. In that case, the processing load required for preprocessing will also increase, but the processing cost (the cost required for building and operating processing resources, etc.) is far greater for the processing performed by the learning model after preprocessing. Therefore, increasing the number of transformation formulas that the general transformation function can express is an effective way to find the optimal transformation formula. 【0088】 For example, the transformation formulas represented by the generalized transformation function may be limited to non-parametric transformation formulas only. The generalized transformation function shown in equation 1 above was capable of representing three transformation formulas, all of which are non-parametric: DFT, DCT-type II, and DHT. In this case, for example, as in the example above, two parameters are used to represent the three transformation formulas. Specifically, it is configured to represent the first transformation formula when α=0, β=0, the second transformation formula when α=1, β=0, and the third transformation formula when α=0, β=1. 【0089】 Similarly, the number of parameters can be reduced to one less than the number of types of transformation formulas. For example, if there are four types of transformation formulas, the first transformation formula can be represented when α=0, β=0, γ=0; the second transformation formula when α=1, β=0, γ=0; the third transformation formula when α=0, β=1, γ=0; and the fourth transformation formula when α=0, β=0, γ=1. Furthermore, if a quantum computer is used as the information processing device 2, it is highly likely that using non-parametric transformation formulas will result in more efficient computation than using parametric transformation formulas. 【0090】 The transformation formula represented by the general transformation function is not limited to this; for example, it may be a transformation formula with parameters. Transformation formulas with parameters include deterministic transformation formulas with parameters (e.g., the Discrete Wavelet Transform (DWT)) or machine learning models with optimized parameters (fixed parameters). 【0091】 To reduce processing costs in preprocessing, it is effective to reduce the number of parameters used in the generalized transformation function. However, if the transformation formula shown by the generalized transformation function includes a parametric transformation formula (such as attention, neural networks, or convolutional neural networks), then multi-head attention is applied to the training data, and attention scores are calculated between many parts, which tends to result in a huge number of calculations (equivalent to the square of the number of tokens, which is the smallest unit when processing the training data). 【0092】 In contrast, if we limit ourselves to non-parametric transformation formulas or transformation formulas with a small number of parameters, as described above, a huge number of calculations like attention scores are unnecessary. Only calculations equivalent to the combinations of parameters included in the general transformation function are required, thus reducing the processing load when performing multiple types of preprocessing. 【0093】 Alternatively, the system may determine suitable parameters for each type of training data and apply parameters that match the data type of the task data. In this case, the information processing device 2, for example, determines the type of training data when it acquires training data in step S1 shown in Figure 4. Data types include, for example, image data, text data, or sound data. The information processing device 2 determines the data type from the file extension of the training data, for example. 【0094】 The information processing device 2 identifies, for example, "jpg," "bmp," "png," and "tiff" as image data, "txt" and "doc" as text data, and "wav," "mp3," and "flac" as audio data. However, since the type of data cannot always be determined by the file extension alone, the information processing device 2 may also use other well-known techniques (techniques for determining the type from header information, techniques using data analysis tools, etc.) to determine the type of data. 【0095】 Then, the information processing device 2 records the optimal parameters determined in S8, associating them with the determined data types. Figure 7 shows an example of recorded parameters. In Figure 7, a parameter database 4 is shown, which stores "data type," "optimal parameters," and "learning model" in a corresponding manner. The information processing device 2 stores the identified data type and the determined optimal parameters in a corresponding manner. The information processing device 2 also stores a learning model, which has undergone machine learning processing on training data that has been preprocessed using a general transformation function to which the optimal parameters are applied, in association with the optimal parameters. 【0096】 In the example in Figure 7, the data type "image" is associated with the optimal parameters "α=0.X1,β=0.X2" and the learning model "Learning Model MDL1". Similarly, the data type "text" is associated with the optimal parameters "α=0.X3,β=0.X4" and the learning model "Learning Model MDL2", and the data type "sound" is associated with the optimal parameters "α=0.X5,β=0.X6" and the learning model "Learning Model MDL3". 【0097】 In this way, the information processing device 2 records the parameters that have been determined to be optimal, associating them with the determined data type. Furthermore, when the information processing device 2 acquires task data in step S11 shown in Figure 6, it determines the data type of the acquired task data. For example, if the task data is image data, the information processing device 2 reads the optimal parameters "α=0.X1,β=0.X2" which are associated with the image data in the parameter database 4 in step S12 shown in Figure 6. 【0098】 Next, in step S14 shown in Figure 6, the information processing device 2 reads the optimized parameters (model parameters) of the learning model MDL1, which are associated with image data, in the parameter database 4. Then, the information processing device 2 performs the processes in steps S15 and S16 and obtains the output that is the result of executing the task. With this configuration, for example, when task data of various data types is input, the degree of task completion can be increased compared to when the data type is not considered. 【0099】 Furthermore, the information processing device 2 may be in an on-premise configuration or a cloud configuration. In the case of a cloud-based information processing device 2, for example, the above-mentioned functions and processing may be provided in the form of SaaS (Software as a Service) or cloud computing. 【0100】 Furthermore, in the above examples, the information processing device 2 performed various storage and control functions, but multiple external devices may be used instead of the information processing device 2. In other words, various information and programs may be distributed and stored across multiple external devices using blockchain technology or the like. 【0101】 Furthermore, the embodiments of this model are not limited to the information processing system 1, but may also be an information processing method or a program. The information processing method may include the steps described above, which are executed by one or more processors. The program may be a program that causes at least one computer to execute the steps described above. 【0102】 The above-mentioned information processing system 1, etc., may be provided in any of the following embodiments. 【0103】 (1) An information processing system comprising one or more processors, wherein in a preprocessing step, the processor preprocesses data using a transformation function, the transformation function is a function having adjustable parameters and representing a transformation expression of a type corresponding to the adjusted parameters from among two or more types of transformation expressions, and in an output step, the processor processes the data that has been preprocessed by the transformation function using machine learning to obtain an output. 【0104】 This configuration allows for easy execution of various pretreatment processes. 【0105】 (2) An information processing system as described in (1) above, wherein the transformation function has a plurality of factors corresponding to each type of transformation formula, and the degree of contribution of the transformation formula to the preprocessing can be adjusted by changing the weight of each of the plurality of factors using the parameter. 【0106】 In this configuration, the contribution of the transformation formula can be easily changed. 【0107】 (3) An information processing system according to (1) or (2) above, wherein the conversion formula is a non-parametric conversion formula or a conversion formula having parameters. 【0108】 This configuration makes it possible to reduce the amount of processing required when performing multiple types of preprocessing. 【0109】 (4) An information processing system according to any one of (1) to (3) above, wherein in the determination step, the processor determines the parameters for which the evaluation result of the learning model optimized by machine learning was good, and the machine learning is machine learning performed using data that has been preprocessed with the conversion function to which the parameters are applied. 【0110】 This approach allows us to identify the parameters that yield high-performance learning models. 【0111】 (5) An information processing system as described in (4) above, wherein in the recording step, the processor records the parameters that have been determined to be good. 【0112】 This configuration makes it easier to reuse parameters that yield high-performance learning models. 【0113】 (6) An information processing system as described in (5) above, wherein in the preprocessing step, the processor preprocesses task data using the conversion function which represents a type of conversion formula corresponding to the recorded good parameters, and in the execution step, the processor inputs the preprocessed task data into a learning model to execute the task, and the learning model is a learning model in which machine learning has been performed on the data that has been preprocessed using the conversion function which represents a type of conversion formula corresponding to the recorded good parameters. 【0114】 This configuration allows for a higher degree of task completion. 【0115】 (7) An information processing method comprising each of the steps described in any one of (1) to (6) above, which is performed by one or more processors. 【0116】 This configuration allows for easy execution of various pretreatment processes. 【0117】 (8) A program that causes a computer to perform any one of the steps described in (1) to (6) above. 【0118】 This configuration allows for easy execution of various pretreatment processes. Of course, this is not always the case. 【0119】 Finally, while various embodiments relating to this disclosure have been described, these are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be implemented in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of symbols] 【0120】 1: Information Processing System 2: Information Processing Device 3: User terminal 4: Parameter Database 23: Processor 33: Processor

Claims

[Claim 1] An information processing system comprising one or more processors, In the preprocessing step, the processor preprocesses the data using a conversion function. The aforementioned conversion function is a function having adjustable parameters, which represents a conversion formula of a type corresponding to the adjusted parameters among two or more types of conversion formulas. In the output step, the processor processes the data that has undergone the preprocessing by the conversion function using machine learning to obtain the output. Information processing system. [Claim 2] In the information processing system described in claim 1, The transformation function has a plurality of factors corresponding to each type of transformation formula, By changing the weight of each of the multiple factors using the aforementioned parameters, the contribution of the transformation formula to the pretreatment can be adjusted. Information processing system. [Claim 3] In the information processing system described in claim 1, The aforementioned transformation formula is a non-parametric transformation formula or a transformation formula with parameters. Information processing system. [Claim 4] In the information processing system described in claim 1, In the determination step, the processor determines the parameters for which the evaluation result of the learning model optimized by machine learning was favorable. The aforementioned machine learning is performed using data that has undergone the preprocessing described above using the transformation function to which the aforementioned parameters are applied. Information processing system. [Claim 5] In the information processing system described in claim 4, In the recording step, the processor records the parameters that have been determined to be good. Information processing system. [Claim 6] In the information processing system described in claim 5, In the preprocessing step, the processor preprocesses the task data using the conversion function that represents a type of conversion formula corresponding to the recorded good parameters. In the execution step, the processor inputs the pre-processed task data into the learning model and executes the task. The learning model is a learning model in which machine learning is performed on data that has been preprocessed by the transformation function that represents a type of transformation formula corresponding to the recorded good parameters. Information processing system. [Claim 7] Information processing method, A process comprising each step according to any one of claims 1 to 6, performed by one or more processors, Information processing methods. [Claim 8] It is a program, The computer is made to perform each of the steps described in any one of claims 1 to 6. program.

Images