Information processing system, information processing method, and information processing program

Abstract

An information processing system according to an embodiment of the present invention comprises at least one processor. The at least one processor acquires, for each of a plurality of component objects, a numeric representation and a composite ratio, executes machine learning on the basis of the plurality of numeric representations to calculate a plurality of regression parameters corresponding to the plurality of component objects, and applies the plurality of composite ratios to a regression model that is defined by the plurality of regression parameters to calculate a predictive value that indicates the characteristics of a composite object obtained by combining the plurality of component objects.

Description

[Technical Field]

[0001] One aspect of the present disclosure relates to an information processing system, an information processing method, and an information processing program. [Background technology]

[0002] A technique using machine learning to analyze a composite object obtained by combining multiple component objects is used. For example, Patent Document 1 describes a method for predicting the binding affinity between the 3D structure of a biopolymer and the 3D structure of a compound. This method includes the steps of generating a predicted 3D structure of a complex of the biopolymer and the compound based on the 3D structures of the biopolymer and the compound, converting the predicted 3D structure into a predicted 3D structure vector representing a result of matching with an interaction pattern, and predicting the binding affinity between the 3D structure of the biopolymer and the 3D structure of the compound by discriminating the predicted 3D structure vector using a machine learning algorithm. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-28879 Summary of the Invention [Problem to be solved by the invention]

[0004] When there are a large number of diverse and diverse component objects, it may be difficult to provide sufficient data for these component objects, and as a result, the accuracy of the analysis of the composite object may not reach the expected level. Therefore, there is a need for a mechanism to improve the accuracy of the analysis of the composite object even when a sufficient amount of data for the component objects is not available. [Means for solving the problem]

[0005] An information processing system according to an aspect of the present disclosure includes at least one processor that acquires a numerical representation and a composite ratio for each of a plurality of component objects, performs machine learning based on the numerical representations to calculate a plurality of regression parameters corresponding to the plurality of component objects, and applies the plurality of composite ratios to a regression model defined by the plurality of regression parameters to calculate a predicted value indicating a characteristic of a composite object obtained by combining the plurality of component objects.

[0006] An information processing method according to an aspect of the present disclosure is executed by an information processing system including at least one processor, and includes the steps of: acquiring a numerical representation and a composite ratio for each of a plurality of component objects; calculating a plurality of regression parameters corresponding to the plurality of component objects by performing machine learning based on the plurality of numerical representations; and calculating a predicted value indicating a characteristic of a composite object obtained by combining the plurality of component objects by applying the plurality of composite ratios to a regression model defined by the plurality of regression parameters.

[0007] An information processing program according to one aspect of the present disclosure causes a computer to execute the steps of obtaining a numerical representation and a composite ratio for each of a plurality of component objects, performing machine learning based on the plurality of numerical representations to calculate a plurality of regression parameters corresponding to the plurality of component objects, and applying the plurality of composite ratios to a regression model defined by the plurality of regression parameters to calculate a predicted value indicating the characteristics of a composite object obtained by combining the plurality of component objects.

[0008] In this aspect, machine learning is performed based on data on each component object to calculate multiple regression parameters corresponding to the multiple component objects. The composite ratio is then applied to a regression model defined by the regression parameters to predict the characteristics of the composite object. By using machine learning and a regression model, it is possible to improve the accuracy of analysis of the composite object even when a sufficient amount of data on the component objects is not available. [Effects of the Invention]

[0009] According to one aspect of the present disclosure, the accuracy of analysis of a composite object can be improved even when a sufficient amount of data is not available for the component objects. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 2 is a diagram illustrating an example of a hardware configuration of a computer that configures the information processing system according to the embodiment. [Figure 2] FIG. 1 is a diagram illustrating an example of a functional configuration of an information processing system according to an embodiment. [Figure 3] 4 is a flowchart illustrating an example of an operation of the information processing system according to the embodiment. [Figure 4] FIG. 10 is a diagram illustrating an example of a procedure for calculating regression parameters. [Figure 5] FIG. 10 is a diagram illustrating another example of a procedure for calculating regression parameters. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same or equivalent elements are designated by the same reference numerals, and redundant description will be omitted.

[0012] [System Overview] The information processing system 10 according to the embodiment is a computer system that performs analysis on a composite object obtained by combining multiple component objects at a given combination ratio. A component object refers to a tangible or intangible entity used to generate the composite object. A composite object can be tangible or intangible. Examples of tangible entities include any substance or object. Examples of intangible entities include data and information. "Combining multiple component objects" refers to the process of converting multiple component objects into a single object, i.e., a composite object. The method of combining is not limited to, and may include, for example, blending, mixing, synthesis, bonding, mixing, merging, combination, compounding, or merging, or other methods. Analysis of a composite object refers to a process for obtaining data that indicates some characteristic of the composite object.

[0013] The multiple-component objects may be any multiple types of materials, in which case the composite object is a multi-component substance produced by those materials. A material is any building block used to produce a multi-component substance. For example, the multiple materials may be any multiple types of molecules or atoms, in which case the composite object is a multi-component substance obtained by combining those molecules or atoms in any manner. For example, the materials may be polymers or monomers, in which case the multi-component substance may be a polymer alloy. The materials may be monomers, in which case the multi-component substance may be a polymer. The materials may be drugs, i.e., chemicals with pharmacological activity, in which case the multi-component substance may be a pharmaceutical.

[0014] The information processing system 10 performs machine learning for analysis of complex objects. Machine learning is a technique for autonomously discovering laws or rules by learning based on given information. The specific technique of machine learning is not limited. For example, the information processing system 10 may perform machine learning using a machine learning model, which is a computational model including a neural network. A neural network is an information processing model that mimics the mechanism of the human nervous system. As a more specific example, the information processing system 10 may perform machine learning using at least one of a graph neural network (GNN), a convolutional neural network (CNN), a recurrent neural network (RNN), an attention RNN, and a multi-head attention.

[0015] [System Configuration] The information processing system 10 is composed of one or more computers. When multiple computers are used, these computers are connected via a communication network such as the Internet or an intranet, thereby logically constructing a single information processing system 10.

[0016] 1 is a diagram showing an example of a general hardware configuration of a computer 100 constituting an information processing system 10. For example, the computer 100 includes a processor 101 such as a CPU that executes an operating system, application programs, etc., a main memory unit 102 consisting of ROM and RAM, an auxiliary memory unit 103 consisting of a hard disk, flash memory, etc., a communication control unit 104 consisting of a network card or wireless communication module, an input device 105 such as a keyboard or mouse, and an output device 106 such as a monitor.

[0017] Each functional element of the information processing system 10 is realized by loading a predetermined program onto the processor 101 or the main memory unit 102 and having the processor 101 execute the program. In accordance with the program, the processor 101 operates the communication control unit 104, the input device 105, or the output device 106, and reads and writes data from and to the main memory unit 102 or the auxiliary memory unit 103. Data or a database required for processing is stored in the main memory unit 102 or the auxiliary memory unit 103.

[0018] 2 is a diagram showing an example of the functional configuration of the information processing system 10. The information processing system 10 includes an acquisition unit 11, a calculation unit 12, and a prediction unit 13 as functional elements.

[0019] The acquisition unit 11 is a functional element that acquires data related to multiple component objects. Specifically, the acquisition unit 11 acquires a numerical representation and a composite ratio for each of the multiple component objects. A numerical representation of a component object refers to data that expresses any attribute of the component object using multiple numerical values. An attribute of a component object refers to a property or characteristic possessed by the component object. The numerical representation may be visualized using various techniques, such as numbers, letters, text, molecular graphs, vectors, images, time-series data, or a combination of any two or more of these techniques. Individual numerical values ​​constituting the numerical representation may be expressed in decimal or other notation systems such as binary or hexadecimal. A composite ratio of component objects refers to the ratio between multiple component objects. The specific type, unit, and expression method of the composite ratio are not limited and may be determined arbitrarily depending on the component object or composite object. For example, the composite ratio may be expressed as a ratio such as a percentage, a histogram, or the absolute amount of each component object.

[0020] The calculation unit 12 is a functional element that calculates regression parameters of a regression model for predicting the characteristics of a composite object. Specifically, the calculation unit 12 calculates the regression parameters by performing machine learning based on multiple numerical expressions corresponding to multiple component objects. A regression model is an equation for determining one or more values ​​of a response variable y when one or more values ​​of an explanatory variable x are given. The regression model may be a linear regression model or a nonlinear regression model. An example of a regression model is the Scheffe polynomial. However, the regression model may also be other parametric models. The regression parameters are numerical values ​​included in the regression model.

[0021] The prediction unit 13 is a functional element that predicts the characteristics of a composite object and outputs the predicted value. The characteristics of a composite object refer to the unique properties of the composite object. Specifically, the prediction unit 13 calculates a predicted value by applying a composite ratio to a regression model defined by the calculated regression parameters. In other words, the prediction unit 13 calculates a predicted value by substituting multiple composite ratios into the regression model.

[0022] In one example, the combination of the calculation unit 12 and the prediction unit 13 is realized by one machine learning model. Alternatively, the calculation unit 12 may be realized by a machine learning model, and the prediction unit 13 may be realized by an algorithm that does not use a machine learning model.

[0023] In one example, each of the at least one machine learning model used in this embodiment is a trained model expected to have the highest estimation accuracy and can therefore be referred to as the "best machine learning model." However, it should be noted that this trained model is not necessarily the "best in reality." A trained model is generated by a given computer processing training data including numerous combinations of input vectors and labels. The given computer inputs input vectors into the machine learning model, calculates an output value, and determines the error between the output value and the label indicated in the training data. The output value is, for example, a predicted value. The error between the output value and the label can be said to be the difference between the estimated result and the correct answer. The computer updates given parameters in the machine learning model based on the error. The computer generates a trained model by repeating this learning process. The computer that generates the trained model is not limited and may be, for example, the information processing system 10 or another computer system. The process of generating a trained model can be referred to as the learning phase, and the process of using the trained model can be referred to as the operation phase.

[0024] In one example, the entire machine learning model used in this embodiment may be described by a function that does not depend on the order of inputs, which can eliminate the influence of the order of multiple vectors in machine learning.

[0025] [System Operation] The operation of the information processing system 10 and the information processing method according to this embodiment will be described with reference to Fig. 3. Fig. 3 is a flowchart showing an example of the operation of the information processing system 10 as a processing flow S1. The processing flow S1 corresponds to the operation phase.

[0026] In step S11, the acquisition unit 11 acquires a numerical representation and a composite ratio for each of the multiple component objects. As an example, if information about two component objects Ea and Eb is input, the acquisition unit 11 acquires, for example, the numerical representation {1,1,2,3,4,3,3,5,6,7,5,4} of the component object Ea, the numerical representation {1,1,5,6,4,3,3,5,1,7,0,0} of the component object Eb, and the composite ratio {0.7, 0.3} of the component objects Ea and Eb. In this example, each numerical representation is represented by a vector. The composite ratio {0.7, 0.3} means that the component objects Ea and Eb are used in a ratio of 7:3 to obtain a composite object.

[0027] The acquisition unit 11 may acquire data for each of the multiple component objects by any method. For example, the acquisition unit 11 may access a given database to read the data, may receive data from another computer or computer system, or may accept data input by a user of the information processing system 10. Alternatively, the acquisition unit 11 may acquire data by any two or more of these methods.

[0028] In step S12, the calculation unit 12 calculates a feature vector for each of the multiple component objects based on its numerical representation. A feature vector is a vector that indicates the characteristics of a component object. A component object's characteristics are any elements that make the component object different from other objects. A vector is an n-dimensional quantity having n numerical values, and can be expressed as a one-dimensional array.

[0029] In step S13, the calculation unit 12 calculates a plurality of regression parameters corresponding to a plurality of component objects based on the calculated plurality of feature vectors.

[0030] In step S14, the prediction unit 13 calculates a predicted value indicating the characteristics of the composite object using a regression model defined by the calculated regression parameters. A regression model defined by regression parameters is, in short, a regression model in which specific, concrete numerical values ​​are determined as regression parameters. The prediction unit 13 applies multiple composite ratios to the regression model to calculate a predicted value.

[0031] In step S15, the prediction unit 13 outputs the predicted value. The method for outputting the predicted value is not limited. For example, the prediction unit 13 may store the predicted value in a given database, transmit it to another computer or computer system, or display it on a display device. Alternatively, the prediction unit 13 may output the predicted value to another functional element for subsequent processing in the information processing system 10.

[0032] The process for the regression model will be described in more detail with reference to Figures 4 and 5. Both Figures 4 and 5 show examples of procedures for calculating regression parameters. In both examples, the component objects represent three materials (polymers): polystyrene, polyacrylic acid, and polybutyl methacrylate. Any form of numerical representation may be provided for each of these materials.

[0033] The Scheffe polynomials are often used in problems related to material blending, so in the examples of Figures 4 and 5, the regression model is the Scheffe polynomial.

[0034] The example of FIG. 4 will be described. In step S121, which is part of step S12, the calculation unit 12 calculates a feature vector Z for each of a plurality of component objects from a numerical expression using a machine learning model for an embedding function for calculating the features of a vector. This machine learning model is a trained model. In an embedding function, an input vector and an output vector have a one-to-one relationship. In this example, the input vector is a numerical expression, and the output vector is a feature vector Z. The calculation unit 12 inputs a plurality of numerical expressions corresponding to a plurality of component objects into the model for the embedding function to calculate the feature vector Z for each of the plurality of component objects. In one example, the calculation unit 12 inputs a numerical expression corresponding to each of the plurality of component objects into the model for the embedding function to calculate the feature vector Z for the component object. In one example, the model for the embedding function may generate a feature vector Z, which is a fixed-length vector, from a numerical expression, which is unstructured data. Unstructured data refers to data that cannot be expressed by a fixed-length vector. In the example of FIG. 4, the calculation unit 12 calculates a feature vector Z1 corresponding to polystyrene, a feature vector Z2 corresponding to polyacrylic acid, and a feature vector Z3 corresponding to polybutyl methacrylate.

[0035] The machine learning model for the embedding function is not limited and may be determined according to any policy taking into account factors such as the types of component objects and composite objects, etc. For example, the calculation unit 12 may execute the embedding function using a graph neural network (GNN), a convolutional neural network (CNN), or a recurrent neural network (RNN).

[0036] In step S122, which is part of step S12, the calculation unit 12 calculates another feature vector M from the feature vector Z for multiple component objects using a machine learning model for an interaction function for interacting multiple vectors. This machine learning model is a trained model. In the interaction function, there is a one-to-one relationship between an input vector and an output vector. In this example, the input vector is the feature vector Z, and the output vector is the feature vector M. In one example, the calculation unit 12 inputs a set of multiple feature vectors Z corresponding to multiple component objects into the model for the interaction function, and calculates a feature vector M for each of the multiple component objects. In the example of FIG. 4, the calculation unit 12 calculates a feature vector M1 corresponding to polystyrene, a feature vector M2 corresponding to polyacrylic acid, and a feature vector M3 corresponding to polybutyl methacrylate.

[0037] The machine learning model for the interaction function is not limited and may be determined according to any policy taking into account factors such as the types of component objects and composite objects. For example, the calculation unit 12 may perform machine learning for the interaction function using an attention RNN or multi-head attention. In another example, the calculation unit 12 may calculate the feature vector M using an interaction function that does not include a learning parameter.

[0038] In step S13 shown in FIG. 4, the calculation unit 12 calculates a regression parameter a of a first-order term of a linear regression model from the feature vector M for each of the multiple component objects. In one example, the calculation unit 12 calculates the regression parameter using a machine learning model. This machine learning model is a trained model. In the function that calculates the regression parameter of the first-order term, there is a one-to-one relationship between the input vector and the output value. In this example, the input vector is the feature vector M, and the output value is the regression parameter a. In one example, the calculation unit 12 inputs a set of multiple feature vectors M corresponding to multiple component objects into the machine learning model and calculates the regression parameter a for each of the multiple component objects. In the example of FIG. 4, the calculation unit 12 calculates a regression parameter a1 corresponding to polystyrene, a regression parameter a2 corresponding to polyacrylic acid, and a regression parameter a3 corresponding to polybutyl methacrylate.

[0039] The machine learning model for calculating the regression parameters is not limited, and may be determined according to any policy taking into consideration factors such as the types of component objects and composite objects, etc. For example, the calculation unit 12 may calculate the regression parameters using a fully connected neural network (FCNN).

[0040] In step S14 shown in FIG. 4, the prediction unit 13 calculates the predicted value E using the following Scheffe polynomial (1), which is defined by three regression parameters a1, a2, and a3. The regression parameter a can also be said to be the regression coefficient of the first-order term in equation (1). The predicted value E indicates the properties of a multi-component material (polymer alloy) obtained from polystyrene, polyacrylic acid, and polybutyl methacrylate. The variable r in equation (1) means the composite ratio. The composite ratios of polystyrene, polyacrylic acid, and polybutyl methacrylate are represented as r1, r2, and r3, respectively.

number

[0041] The example of Fig. 5 will be described. In the example of Fig. 5, step S12 including step S121 and step S122 is the same as the example of Fig. 4, but steps S13 and S14 are different from the example of Fig. 4.

[0042] In step S13 shown in FIG. 5, the calculation unit 12 calculates regression parameters of a linear regression model from the feature vector M for each of the multiple component objects. Specifically, the calculation unit 12 calculates a regression parameter a for the linear term and a regression parameter b for the quadratic term. In one example, the calculation unit 12 calculates the regression parameters using machine learning such as FCNN. A machine learning model is prepared for each of the linear term and quadratic term of the linear regression model.

[0043] As in the example of FIG. 4, in the function that calculates the regression parameter of the linear term, the input vector and the output value have a one-to-one relationship. In this example, the input vector is a feature vector M, and the output value is a regression parameter a. In one example, the calculation unit 12 inputs a set of multiple feature vectors M corresponding to multiple component objects into a machine learning model and calculates the regression parameter a for each of the multiple component objects. In the example of FIG. 5, the calculation unit 12 also calculates a regression parameter a1 corresponding to polystyrene, a regression parameter a2 corresponding to polyacrylic acid, and a regression parameter a3 corresponding to polybutyl methacrylate.

[0044] In a function for calculating a quadratic regression parameter, each input vector is obtained by combining two feature vectors. This function calculates one regression parameter from two vectors. In this example, two feature vectors M are combined. In the example of FIG. 5, the calculation unit 12 combines two feature vectors M1 and M2 to generate a first input vector, combines two feature vectors M1 and M3 to generate a second input vector, and combines two feature vectors M2 and M3 to generate a third input vector. Therefore, the first input vector corresponds to polystyrene and polyacrylic acid, the second input vector corresponds to polystyrene and polybutyl methacrylate, and the third input vector corresponds to polyacrylic acid and polybutyl methacrylate. In a function for calculating a quadratic regression parameter, there is also a one-to-one relationship between the input vector and the output value. In this example, the input vector is a combination of two feature vectors M, and the output value is the regression parameter b. In one example, the calculation unit 12 inputs all combinations of input vectors into a machine learning model and calculates the regression parameter b for each combination. In the example of FIG. 5, the calculation unit 12 calculates the regression parameter b corresponding to the combination of polystyrene and polyacrylic acid. 12 and the regression parameter b corresponding to the combination of polystyrene and polybutyl methacrylate. 13 and the regression parameter b corresponding to the combination of polyacrylic acid and polybutyl methacrylate. 23 and calculate.

[0045] In step S14 shown in FIG. 5, the prediction unit 13 calculates six regression parameters a1, a2, a3, b 12 ,b 13 ,b 23 The predicted value E is calculated using the following Scheffe polynomial (2) defined by: In equation (2), the regression parameter a can be said to be the regression coefficient of the linear term, and the regression parameter b can be said to be the regression coefficient of the quadratic term. The variable r in equation (2) means a composite ratio, just like in equation (1).

number

[0046] Although three component objects are shown in FIGS. 4 and 5, the number of component objects is of course not limited, and information processing system 10 may process any number of component objects.

[0047] Similarly, for a regression model including third-order or higher terms or other parameters, the information processing system 10 may output individual regression parameters based on the feature vectors of all related component objects. When calculating a regression parameter that does not depend on a specific explanatory variable, such as the intercept of a linear regression, the information processing system 10 may output a single regression parameter based on the feature vectors of all component objects.

[0048] 4 and 5, the calculation unit 12 executes both the embedding function and the interaction function, but one of these two functions may be omitted. For example, the calculation unit 12 may calculate the regression parameters from the feature vector Z obtained by a machine learning model for the embedding function. In either case, the calculation unit 12 executes machine learning to calculate the regression parameters.

[0049] In one example, the machine learning model for the embedding function, the machine learning model for the interaction function, the machine learning model for the regression parameters, and the regression model may be constructed by a single neural network or a collection of multiple neural networks. Alternatively, the machine learning model for the embedding function, the machine learning model for the interaction function, and the machine learning model for the regression parameters may be constructed by a single neural network or a collection of multiple neural networks.

[0050] [program] An information processing program for causing a computer or computer system to function as the information processing system 10 includes program code for causing the computer system to function as the acquisition unit 11, the calculation unit 12, and the prediction unit 13. This information processing program may be provided by being non-temporarily recorded on a tangible recording medium such as a CD-ROM, a DVD-ROM, or a semiconductor memory. Alternatively, the information processing program may be provided via a communication network as a data signal superimposed on a carrier wave. The provided information processing program is stored in, for example, the auxiliary storage unit 103. The processor 101 reads out and executes the information processing program from the auxiliary storage unit 103, thereby realizing each of the above-mentioned functional elements.

[0051] [effect] As described above, an information processing system according to one aspect of the present disclosure includes at least one processor that acquires a numerical representation and a composite ratio for each of a plurality of component objects, performs machine learning based on the numerical representations to calculate a plurality of regression parameters corresponding to the plurality of component objects, and applies the plurality of composite ratios to a regression model defined by the plurality of regression parameters to calculate a predicted value indicating a characteristic of a composite object obtained by combining the plurality of component objects.

[0052] An information processing method according to an aspect of the present disclosure is executed by an information processing system including at least one processor, and includes the steps of: acquiring a numerical representation and a composite ratio for each of a plurality of component objects; calculating a plurality of regression parameters corresponding to the plurality of component objects by performing machine learning based on the plurality of numerical representations; and calculating a predicted value indicating a characteristic of a composite object obtained by combining the plurality of component objects by applying the plurality of composite ratios to a regression model defined by the plurality of regression parameters.

[0053] An information processing program according to one aspect of the present disclosure causes a computer to execute the steps of obtaining a numerical representation and a composite ratio for each of a plurality of component objects, performing machine learning based on the plurality of numerical representations to calculate a plurality of regression parameters corresponding to the plurality of component objects, and applying the plurality of composite ratios to a regression model defined by the plurality of regression parameters to calculate a predicted value indicating the characteristics of a composite object obtained by combining the plurality of component objects.

[0054] In this aspect, machine learning is performed based on data on each component object to calculate multiple regression parameters corresponding to the multiple component objects. The composite ratio is then applied to a regression model defined by the regression parameters to predict the characteristics of the composite object. By using machine learning and a regression model, it is possible to improve the accuracy of analysis of the composite object even when a sufficient amount of data on the component objects is not available.

[0055] Once the regression parameters are obtained, the composite ratio can be changed and the properties of the composite object can be instantly recalculated using the regression model. In other words, the calculated regression parameters can be reused. By adopting a method for estimating regression parameters using machine learning, it becomes possible to quickly search for the properties of a composite object while changing the composite ratio.

[0056] In another aspect of the information processing system, at least one processor may input a plurality of numerical expressions to a first machine learning model to calculate a plurality of feature vectors corresponding to a plurality of component objects, and input the plurality of feature vectors to a second machine learning model to calculate a plurality of regression parameters. This series of steps can further improve the accuracy of analysis of complex objects even when a sufficient amount of data about the component objects is not available.

[0057] In another aspect of the information processing system, the first machine learning model may include a machine learning model for an embedding function and a machine learning model for an interaction function. The at least one processor may input a plurality of numerical expressions to the machine learning model for the embedding function to calculate a plurality of first feature vectors corresponding to a plurality of component objects, input the plurality of first feature vectors to the machine learning model for the interaction function to calculate a plurality of second feature vectors corresponding to the plurality of component objects, and input the plurality of second feature vectors to the second machine learning model to calculate a plurality of regression parameters. Configuring the first machine learning model in this manner further improves the accuracy of analysis of complex objects even when a sufficient amount of data on the component objects is not available.

[0058] In another aspect of the information processing system, the machine learning model for the embedding function may be a machine learning model that generates a first feature vector, which is a fixed-length vector, from a numerical expression that is unstructured data. By using this machine learning model, a feature vector can be obtained from a numerical expression that cannot be expressed by a fixed-length vector.

[0059] In an information processing system according to another aspect, the regression model may be a Scheffe polynomial. At least one processor may calculate, as the multiple regression parameters, multiple regression coefficients of the first-order term of the Scheffe polynomial. By using the Scheffe polynomial, which is often used in blending problems, it is possible to accurately analyze a composite object obtained by blending multiple component objects. In addition, the regression coefficient of the first-order term can be used to calculate a predicted value that takes into account the individual influence of each component object.

[0060] In another aspect of the information processing system, the at least one processor may further calculate a plurality of regression coefficients of quadratic terms of Scheffe polynomials as the plurality of regression parameters. In this case, the regression coefficients of the quadratic terms can be used to calculate a predicted value that further takes into account the influence of a combination of two component objects.

[0061] In another aspect of the information processing system, the component object may be a material, and the composite object may be a multi-component substance, which can improve the accuracy of analysis of the multi-component substance even when a sufficient amount of data about the material is not available.

[0062] In an information processing system according to another aspect, the material may be a polymer or a monomer, and the multi-component substance may be a polymer alloy. In this case, it is possible to improve the accuracy of analysis of the polymer alloy even when a sufficient amount of data on the polymer or monomer is not available. There is a great diversity of polymers and monomers, and correspondingly, the number of types of polymer alloys is enormous. For such polymers, monomers, and polymer alloys, experiments can generally be performed on only a portion of the possible combinations, and therefore, a sufficient amount of data is often not obtained. According to this aspect, it is possible to accurately analyze the polymer alloy even when such data is insufficient.

[0063] [Variation] The present invention has been described in detail above based on the embodiments. However, the present invention is not limited to the above embodiments. Various modifications of the present invention are possible without departing from the spirit and scope of the present invention.

[0064] The processing procedure of the information processing method executed by at least one processor is not limited to the example in the above embodiment. For example, some of the steps or processes described above may be omitted, or the steps may be executed in a different order. Furthermore, any two or more of the steps described above may be combined, or some of the steps may be modified or deleted. Alternatively, other steps may be executed in addition to the steps described above.

[0065] When comparing the magnitude of two numbers in an information processing system, either of the two criteria "greater than or equal to" and "greater than" may be used, or either of the two criteria "less than or equal to" and "under." The selection of such criteria does not change the technical significance of the process of comparing the magnitude of two numbers.

[0066] In this disclosure, the expression "at least one processor executes a first process, executes a second process, ... executes an nth process" or a corresponding expression indicates a concept including a case where the processor that executes n processes from the first process to the nth process changes midway. In other words, this expression indicates a concept including both a case where all n processes are executed by the same processor and a case where the processor changes among the n processes according to an arbitrary policy. [Explanation of symbols]

[0067] 10...information processing system, 11...acquisition unit, 12...calculation unit, 13...prediction unit

Claims

1. at least one processor; the at least one processor: obtaining a numerical representation and a composite ratio for each of the plurality of component objects; inputting the plurality of numerical representations into a first machine learning model to calculate a plurality of feature vectors corresponding to the plurality of component objects; inputting the plurality of feature vectors into a second machine learning model to calculate a plurality of regression parameters corresponding to the plurality of component objects; applying the plurality of composite ratios to a regression model defined by the plurality of regression parameters to calculate a predicted value indicative of a property of a composite object obtained by combining the plurality of component objects; Information processing system.

2. the first machine learning model includes a machine learning model for an embedding function and a machine learning model for an interaction function; the at least one processor: inputting the plurality of numerical representations into a machine learning model for the embedding function to calculate a plurality of first feature vectors corresponding to the plurality of component objects; inputting the plurality of first feature vectors into a machine learning model for the interaction function to calculate a plurality of second feature vectors corresponding to the plurality of component objects; inputting the plurality of second feature vectors into the second machine learning model to calculate the plurality of regression parameters; The information processing system according to claim 1 .

3. the machine learning model for the embedding function is a machine learning model that generates the first feature vector, which is a fixed-length vector, from the numerical expression, which is unstructured data; The information processing system according to claim 2 .

4. the regression model is a Scheffe polynomial; the at least one processor calculates a plurality of regression coefficients of first-order terms of the Scheffe polynomial as the plurality of regression parameters; The information processing system according to any one of claims 1 to 3.

5. the at least one processor further calculates a plurality of regression coefficients of a quadratic term of the Scheffe polynomial as the plurality of regression parameters. The information processing system according to claim 4 .

6. the component object is a material and the composite object is a multi-component substance; The information processing system according to any one of claims 1 to 5.

7. the material is a polymer or a monomer, and the multi-component substance is a polymer alloy; The information processing system according to claim 6.

8. 1. An information processing method executed by an information processing system including at least one processor, comprising: obtaining a numerical representation and a composite ratio for each of a plurality of component objects; inputting the plurality of numerical representations into a first machine learning model to calculate a plurality of feature vectors corresponding to the plurality of component objects; inputting the plurality of feature vectors into a second machine learning model to calculate a plurality of regression parameters corresponding to the plurality of component objects; applying the composite ratios to a regression model defined by the regression parameters to calculate a predicted value indicative of a property of a composite object obtained by combining the component objects; An information processing method including:

9. obtaining a numerical representation and a composite ratio for each of a plurality of component objects; inputting the plurality of numerical representations into a first machine learning model to calculate a plurality of feature vectors corresponding to the plurality of component objects; inputting the plurality of feature vectors into a second machine learning model to calculate a plurality of regression parameters corresponding to the plurality of component objects; applying the composite ratios to a regression model defined by the regression parameters to calculate a predicted value indicative of a property of a composite object obtained by combining the component objects; An information processing program that causes a computer to execute the above.

Images