Information processing device, information processing method, program, composition and method for producing the same

Abstract

The present invention provides an information processing device, method, and program for accurately predicting the reaction rate when a compound or composition is irradiated with light, as well as a composition and a method for producing the same. [Solution] In an information processing system in which an information processing device and a user device are connected via a network, the information processing device 100 has a processor including a first acquisition unit that acquires a learning dataset including information on the training compound, information on the curing conditions, and information on the reaction rate, and a learning unit that constructs a learning model based on the training dataset, and the learning unit constructs the learning model after evaluating the performance of candidate models by double cross validation, and in cross validation outside the double cross validation, the training dataset is divided into k folds for each training compound, and k-1 (where k is any natural number and k>1) folds are used as first training data and one fold is used as first test data.

Description

[Technical Field]

[0001] The present invention relates to an information processing device, an information processing method, a program, a composition, and a method for producing the same. [Background technology]

[0002] In recent years, developments have been underway in materials informatics, which uses techniques such as machine learning to predict material properties and optimize experimental processes (see, for example, Patent Documents 1 to 3). Materials informatics uses machine learning algorithms and artificial intelligence (AI) technology to learn patterns in materials data and predict new material candidates and physical properties. This is expected to lead to faster material development than conventional experimental approaches. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] WO2021 / 200281 [Patent Document 2] Japanese Patent Application Publication No. 2024-078744 [Patent Document 3] WO2023 / 032809 Summary of the Invention [Problem to be solved by the invention]

[0004] Curable resin compositions are widely used in industrial products as insulating films that cover the surface of printed circuit boards and protect circuit patterns, and as materials used to form circuits on boards. Therefore, there is a need for a technology that can accurately predict the reaction rate when such curable compounds or compositions are cured. The technologies described in Patent Documents 1 and 2 are not intended for curable compounds or compositions. Furthermore, the technology described in Patent Document 3 is also applicable to curable compounds or compositions, but in order to improve prediction reliability, it is necessary to prepare a training dataset with a sufficient number of data, and the burden of experiments required to prepare a sufficient number of data remains unresolved.

[0005] The present invention has been made in consideration of the above-mentioned problems, and aims to provide an information processing device, an information processing method, and a program suitable for predicting the reaction rate when a compound or composition is cured even when the number of data used to create a learning model is small, as well as a composition and a method for producing the same. [Means for solving the problem]

[0006] An information processing device according to one embodiment of the present invention comprises a first acquisition unit that acquires a training dataset including information on training compounds, information on curing conditions, and information on reaction rates, and a learning unit that constructs a training model based on the training dataset, wherein the learning unit constructs the training model through performance evaluation of candidate models by double cross-validation, and in cross-validation outside the double cross-validation, the training dataset is divided into k folds for each training compound, and k-1 (where k is any natural number, k>1) folds are used as first training data and one fold is used as first test data.

[0007] An information processing device according to one aspect of the present invention further includes a second acquisition unit that acquires information about predicted compounds, and an output unit that outputs information about the predicted compounds and information about the reaction rates of the predicted compounds based on the learning model.

[0008] An information processing method according to one aspect of the present invention includes an information processing device that executes the steps of acquiring a training dataset including information on a training compound, information on curing conditions, and information on a reaction rate, and constructing a training model based on the training dataset. The step of constructing the training model involves constructing the training model through performance evaluation of candidate models by double cross-validation, and in cross-validation outside the double cross-validation, dividing the training dataset into k folds for each training compound, with k-1 folds (where k is any natural number and k>1) being used as first training data and one fold being used as first test data.

[0009] In an information processing method according to one aspect of the present invention, the information processing device further executes a step of acquiring information about a predicted compound, and a step of outputting information about the reaction rate of the predicted compound based on the information about the predicted compound and the learning model.

[0010] A program according to one aspect of the present invention causes an information processing device to execute the steps of acquiring a training dataset including information on training compounds, information on curing conditions, and information on reaction rates, and constructing a training model based on the training dataset, wherein the step of constructing the training model involves constructing the training model through performance evaluation of candidate models by double cross-validation, and in cross-validation outside the double cross-validation, the training dataset is divided into k folds for each training compound, and k-1 (where k is any natural number, k>1) folds are used as first training data and one fold is used as first test data.

[0011] A composition according to one aspect of the present invention includes a compound selected based on information about the predicted reaction rate of the compound output by the information processing device or the information processing method.

[0012] A method for producing a composition according to one embodiment of the present invention includes a step of blending a compound selected based on information regarding the predicted reaction rate of the compound output by the information processing device or the information processing method into a composition. [Effects of the Invention]

[0013] According to the present invention, it is possible to provide an appropriate information processing device, information processing method, and program for predicting the reaction rate when a compound or composition is cured, even when the number of data used to create a learning model is small, as well as a composition and a method for manufacturing the same. [Brief explanation of the drawings]

[0014] [Figure 1A] 1 is a schematic diagram illustrating a configuration of an information processing system according to an embodiment of the present invention. [Figure 1B] 1 is a schematic diagram illustrating a hardware configuration and a functional configuration of an information processing apparatus according to an embodiment of the present invention. [Figure 2A] FIG. 2 is a schematic cross-sectional view of a coating film of the composition being irradiated with light. [Figure 2B] 1 is a graph showing the reaction rate when compositions using compounds A to H as monomers and the same photopolymerization initiator are irradiated with a predetermined amount of ultraviolet light. [Figure 2C] FIG. 1 is a schematic diagram showing an image of a typical division method. [Figure 2D] FIG. 1 is a schematic diagram illustrating an image of a division method according to the present embodiment. [Figure 2E] FIG. 1 is a diagram illustrating an example of a training dataset. [Figure 3] 1 is a flowchart of an information processing method according to an embodiment of the present invention. [Figure 4A] FIG. 1 is a diagram showing predicted values ​​of the relationship between light irradiation intensity and reaction rate of predicted compounds obtained in the Examples. [Figure 4B] FIG. 1 is a graph showing the measured values ​​of the relationship between the light irradiation intensity and the reaction rate of a composition using a predicted compound obtained in an example. DETAILED DESCRIPTION OF THE INVENTION

[0015] Below, we will explain in detail an embodiment of the present invention (hereinafter referred to as the "present embodiment") with reference to the drawings, but the present invention is not limited to this and various modifications are possible within the scope of the gist of the present invention.

[0016] 1. Information processing equipment 1A is a schematic diagram showing the configuration of an information processing system 1 according to one embodiment of the present invention. As shown in FIG. 1A, in this example of the information processing system 1, a server 100 (hereinafter also referred to as "information processing device 100") serving as an information processing device and a user device 200 are communicably connected via a network N such as the Internet.

[0017] The information processing device 100 is an information processing device realized by a program, and may transmit processing results to the user device 200 via the communication interface 120 and the network N in response to a processing request received from the user device 200. For example, the information processing device 100 acquires a training dataset from the user device 200, including information on training compounds, information on curing conditions, and information on reaction rates. Then, the information processing device 100 constructs a training model based on the training dataset. Furthermore, the information processing device 100 may acquire information on predicted compounds, calculate information on the reaction rates of the predicted compounds based on the information on the predicted compounds and the training model, and transmit the information to the user device 200.

[0018] The user device 200 is an information processing device used by a user to execute information processing, and may be, for example, a computer, a smartphone, a tablet terminal, a personal computer, or the like.

[0019] Note that Figure 1A shows a client / server system including an information processing device 100 and a user device 200, and the following describes a mode in which the server functions as the information processing device 100, but the system of this embodiment is not limited to this, and instead of this system configuration, the user device 200 may be equipped with the processing functions of the information processing device described below.

[0020] Hereinafter, the hardware configuration and functional configuration of the information processing device 100 will be described with reference to FIG. 1B, and then each control will be described in detail in association with the functional configuration of the information processing device 100.

[0021] As shown in FIG. 1B, the information processing device 100 includes, for example, a processor 110, a communication interface 120, an input / output interface 130, a memory 140, a storage 150, and one or more communication buses 160 for interconnecting these components.

[0022] The processor 110 executes processes, functions, or methods implemented by code or instructions included in a program stored in the storage 150. The processor 110 may include, for example and without limitation, one or more central processing units (CPUs), MPUs, GPUs, etc., and may implement each process, function, or method disclosed in each embodiment by a logic circuit (hardware) formed in an integrated circuit or the like, or a dedicated circuit.

[0023] As shown in FIG. 1B, the processor 110 of this embodiment may be configured to function as a first acquisition unit 111, a learning unit 112, a second acquisition unit 113, and an output unit 114.

[0024] The communication interface 120 transmits and receives various data to and from other devices via the network N. The communication may be performed either wired or wirelessly, and any communication protocol may be used as long as mutual communication is possible. For example, the communication interface 120 may be implemented as hardware such as a network adapter, various communication software, or a combination of these.

[0025] Network N may be, by way of example and not limitation, an ad-hoc network, an intranet, an extranet, a virtual private network (VPN), a local area network (LAN), a wireless LAN (WLAN), a wide area network (WAN), a wireless WAN (WWAN), a metropolitan area network (MAN), a portion of the Internet, a portion of the public switched telephone network (PSTN), a cellular network, Integrated Service Digital Networks (ISDNs), wireless LANs, Long Term Evolution (LTE), Code Division Multiple Access (CDMA), Bluetooth, satellite communications, etc., or any combination thereof. A network may include one or more networks.

[0026] The input / output interface 130 includes an input device for inputting various operations to the information processing device 100, and an output device for outputting processing results processed by the information processing device 100. For example, the input / output interface 130 includes information input devices such as a keyboard, a mouse, and a touch panel, and information output devices such as a display. Note that the information processing device 100 may receive a predetermined input or execute a predetermined output by connecting an external input / output interface 130.

[0027] Memory 140 temporarily stores programs loaded from storage 150 and provides a working area for processor 110. Memory 140 also temporarily stores various data generated while processor 110 is executing the programs. Memory 140 may be, for example, a high-speed random access memory such as a DRAM, an SRAM, a DDR RAM, or another random access solid-state storage device, or a combination of these.

[0028] Storage 150 stores programs, various functional units, and various data. Storage 150 may be, for example, one or more magnetic disk storage devices, optical disk storage devices, flash memory devices, or nonvolatile memories such as other nonvolatile solid-state storage devices, or a combination thereof. Another example of storage 150 may be one or more storage devices installed remotely from processor 110.

[0029] The communication bus 160 is not particularly limited as long as it is a known dedicated communication path for exchanging data, control information, and the like between hardware configurations.

[0030] Next, each functional unit of the information processing device of this embodiment will be described in detail. Before going into the details of this embodiment, however, as a premise, a curable compound or composition will be described using a photosensitive composition as an example.

[0031] FIG. 2A shows a schematic cross-sectional view of a coating film 10 formed on a substrate 30 and made of a composition containing a monomer 11 and a photopolymerization initiator 12, irradiated with light such as ultraviolet light from a light source 20. As shown in FIG. 2A, when light is irradiated toward the coating film 10, the light causes the photopolymerization initiator 12 to generate radicals, initiating radical polymerization. The radicals generated from the photopolymerization initiator 12 react with the polymerizable double bond of the monomer 11 to generate new radicals, which then react with the polymerizable double bond of another monomer 11 to generate new radicals. This process continues, resulting in a continuous reaction and the formation of a polymer. Furthermore, in the case of a polyfunctional monomer, crosslinking occurs and the polymer hardens.

[0032] Figure 2B shows the reaction rate when a predetermined amount of UV light is irradiated to a composition using compounds A to H as monomers and the same photopolymerization initiator. The horizontal axis of Figure 2B represents the UV irradiation dose, and the vertical axis represents the reaction rate. As shown in Figure 2B, the reaction rate relative to the UV irradiation dose varies depending on the monomer compound. Furthermore, the reaction rate varies depending on the type of compound and the irradiation dose, with some compounds showing no increase in reaction rate above a certain irradiation dose and others showing a reaction rate that increases in proportion to the irradiation dose. The reaction rate was calculated from the disappearance / remaining amount of the peak derived from the polymerizable functional group of the monomer before and after UV irradiation using infrared spectroscopy (IR). The calculation method for the reaction rate is not limited to this; various reaction rate calculation methods can be used depending on the type of monomer and polymerizable functional group.

[0033] Furthermore, although not shown, when a composition containing different types of photopolymerization initiators and the same monomers is irradiated with a predetermined amount of ultraviolet light, the reaction rate varies depending on the type of photopolymerization initiator and the amount of irradiation.

[0034] As shown in Figure 2B, the reaction rate is not 100%. There are various reasons why the reaction rate does not reach 100%, one of which is thought to be the cessation of the reaction due to the disappearance of radicals. Examples of ways in which the reaction stops due to the disappearance of radicals include the annihilation of radicals themselves or the deactivation of radicals by oxygen or the like. The reaction rate does not necessarily have to be 100%, but if the reaction stops before the reaction rate is sufficiently high, the coating film will not cure sufficiently, which can lead to inconveniences such as not being able to achieve the desired physical properties.

[0035] As can be seen from Figure 2B, the reaction rate can vary depending on the combination of monomers and photopolymerization initiators used, and confirming such a reaction rate through an experimental process can take several days. Therefore, if the reaction rate could be obtained from information about the compound and information about the curing conditions (light irradiation conditions) without going through an experimental process, it would be possible to efficiently investigate the composition of photosensitive compounds or compositions.

[0036] The same applies to thermosetting compounds or compositions as to the photosensitive compounds or compositions described above. That is, in thermosetting compounds or compositions, the reaction rate can vary depending on various conditions, such as the combination of monomers and curing agents used, and heating conditions such as heating temperature and heating time. Therefore, it is necessary to obtain the reaction rate from information about the thermosetting compound and information about the curing conditions (heating conditions).

[0037] In the following description of this embodiment, a photosensitive compound or composition will be taken as an example, but the present invention is also applicable to a thermosetting compound or composition.

[0038] In this embodiment, the first acquisition unit 111 acquires a training data set including information on a training compound, information on curing conditions, and information on a reaction rate. Here, the training compound is not particularly limited, but may be, for example, a monomer and / or a photopolymerization initiator.

[0039] The information about the training compound is not particularly limited, but may include, for example, information about the positions of atoms and bond types in the molecules of a monomer or a photopolymerization initiator. Examples of data containing such information include MOL files and SMILES (Simplified Molecular Input Line Entry System). The chemical structure data may be converted into structural descriptors using any library such as RDKit, mordred, or fingerprints.

[0040] The information about the training compound may not only include information about a single compound, but also information about the composition, such as multiple compounds contained in the composition and their content ratios, if the target is a composition. For example, if the training compound includes a monomer and a photopolymerization initiator, the information about the training compound may include information about the monomer and the photopolymerization initiator. Furthermore, if the training compound includes two or more monomers, the information about the training compound may include information indicating the molecular structure, such as SMILES, and information about the content ratio of each monomer. Alternatively, if the training compound includes other components, such as a leveling agent or filler, in addition to the monomer and the photopolymerization initiator, it may include information about the compounds of the other components.

[0041] The information on the curing conditions is not particularly limited, and examples thereof include information on the light irradiation intensity, irradiation time, wavelength, and irradiation atmosphere. Information on the irradiation atmosphere includes information on the atmosphere in which light irradiation was performed, such as normal air containing oxygen or an inert gas atmosphere such as nitrogen. In the case of a thermosetting compound or composition, information on the heating temperature, heating time, and heating atmosphere includes, for example, information on the heating atmosphere.

[0042] The information on the reaction rate includes information on the training compound and information on the reaction rate corresponding to the information on the curing conditions. That is, the training dataset is used as a dataset in which the information on the training compound and the information on the curing conditions are associated with the information on the reaction rate.

[0043] The learning unit 112 performs performance evaluation of the candidate model by double cross validation, and then constructs a learning model to be used by the output unit 114. In this case, in this embodiment, in cross validation outside the double cross validation, the learning dataset is divided into k folds for each learning compound, and k-1 (where k is any natural number, k>1) folds are used as first training data, and one fold is used as first test data.

[0044] In this embodiment, by dividing the training data set for each training compound, it becomes possible to appropriately evaluate the training model, and as a result, it becomes possible to create an appropriate training model for predicting the reaction rate when a compound or composition is cured. This point will be described below.

[0045] First, before describing the training dataset division method of this embodiment, a typical double cross-validation division method will be described. In typical cross-validation outside of double cross-validation, the training dataset is randomly divided into k folds, and training data and test data are created from the divided folds. Alternatively, when the dataset is particularly small, each data point in the dataset is used as a test set, and all other data points are used as a training set (Leave-One-Out). Using the double cross-validation method, multiple combinations of training data and test data can be created from a single dataset, allowing for appropriate verification of prediction results even with a small amount of data, and enabling the creation of a learning model using an appropriate preprocessing method and regression analysis technique. In a typical division method, the training dataset is randomly divided to eliminate data bias and improve the reliability of prediction results.

[0046] Figure 2C shows an image of a typical division method. As shown in the graph in Figure 2B, the training dataset includes data on the same training compound A, in which the reaction rate is measured after irradiation under different light irradiation conditions. Therefore, as shown in Figure 2C, the training dataset may contain multiple pieces of data on the same training compound, such as training compound A. In Figure 2C, the training data is shown in a white frame, and the test data is shown in a gray frame.

[0047] In this situation, if the training dataset is randomly divided as shown in Figure 2C, both the training data and the test data will contain information about training compound B, resulting in a situation where a learning model created with training data containing information about training compound B is verified using test data that also contain information about training compound B. If a learning model created with training data containing information about training compound B is used to predict test data containing information about training compound B, the accuracy will naturally be high, so the predictive accuracy of the learning model may be overestimated. Using a learning model constructed with such overestimation may reduce the reliability of reaction rate prediction results for novel compounds, making it unsuitable for practical use.

[0048] Therefore, in this embodiment, as shown in FIG. 2D, the training dataset is divided into k folds for each training compound, with k-1 folds (where k is any natural number and k>1) serving as the first training data and one fold serving as the first test data. This prevents both the training data and the test data from containing information about the same training compound. Therefore, overestimation of the predictive accuracy of the learning model can be suppressed, and by constructing a learning model with an appropriately evaluated predictive accuracy, it becomes possible to accurately predict reaction rates for novel compounds.

[0049] In the double cross-validation of this embodiment, the first training data may be randomly divided into m folds in the inner cross-validation, and m-1 (where m is any natural number and m>1) folds may be used as the inner training data, and one fold may be used as the inner test data. This validation may be repeated m times to determine the hyperparameters. Leave-one-out may also be used, and the number of folds may be selected arbitrarily, such as 3 folds or 5 folds.

[0050] Then, in the outer cross-validation, a learning model candidate is constructed using the first training data using the hyperparameters determined in the inner cross-validation, and the learning model candidate is verified using the first test data, repeating this process k times for performance evaluation, and a learning model can be constructed based on the performance evaluation.

[0051] By using this double cross-validation method, it is possible to determine appropriate regression analysis methods and preprocessing methods even when the number of data in the training dataset is small, and to construct an appropriate learning model for predicting response rates.

[0052] There are no particular restrictions on the method used to build the learning model, and PLS (partial least squares), Ridge, LASSO (least absolute shrinkage and selection operator) are also available. Linear regression analysis methods such as Elastic Net, linear SVR (support vector regression), and linear GPR (Gaussian process regression) can be used; nonlinear regression analysis methods such as nonlinear SVR, Random Forest, XGBoost (extreme gradient boosting), LightGBM (light gradient boosting model), nonlinear GPR, and neural network can be used. In addition, any variable selection (feature selection) method such as Boruta can be applied.

[0053] Furthermore, the explanatory variables and the objective variables of the training dataset may be transformed. An example of a training dataset is shown in FIG. 2E. The training dataset shown in FIG. 2E includes information on compounds serving as explanatory variables, information on curing conditions, and information on the reaction rate serving as the objective variable (such as experimental data records). For example, information on curing conditions serving as explanatory variables (exposure dose) may be transformed into log, √, or reciprocal. As mentioned above, information on compounds serving as explanatory variables may be transformed using any library such as RDKit, Mordred, or fingerprint.

[0054] Next, the output process in the information processing device and information processing method of this embodiment will be described with reference to FIG.

[0055] In step S01, the second acquisition unit 113 receives input from a user via the input / output interface 130 or the like and acquires information about the predicted compound. At this time, the second acquisition unit 113 may also acquire information about the curing conditions.

[0056] The information about the predicted compound is the same as the information about the training compound described above. The prediction compound may be a compound different from the training compound. As described above, the information processing device and information processing method of this embodiment suppress overestimation of the prediction accuracy of the training model and construct a training model with an appropriately evaluated prediction accuracy, thereby making it possible to appropriately predict reaction rates even for compounds different from the training compound.

[0057] Then, in step S02, the output unit 114 outputs information about the reaction rate of the predicted compound using the learning model described above, based on the information about the predicted compound and the information about the curing conditions. The information about the reaction rate of the predicted compound may be the relationship between the curing conditions and the reaction rate of the predicted compound, or may be the reaction rate of the predicted compound under specific curing conditions. The information about the curing conditions may be acquired by the second acquisition unit 113 through user input, or may be acquired as an arbitrary value by the information processing device.

[0058] In this manner, in this embodiment, by using a predetermined learning model, it is possible to properly predict information regarding the reaction rate of a predicted compound based on information regarding the predicted compound.

[0059] Furthermore, the output unit 114 of the information processing device of this embodiment may output information about a compound and / or information about curing conditions that satisfy the information about the target reaction rate, based on information about the target reaction rate targeted by the user. For example, the output unit 114 may repeat steps S01 and S02 while changing the information about the predicted compound, and when the reaction rate of the predicted compound becomes equal to or greater than the reaction rate targeted by the user, output information about the predicted compound.

[0060] 2. Information Processing Method In the information processing method of this embodiment, an information processing device executes the steps of acquiring a training dataset including information on training compounds, information on curing conditions, and information on reaction rates, and constructing a training model based on the training dataset. The step of constructing the training model involves constructing the training model through performance evaluation of candidate models by double cross-validation, and in cross-validation outside the double cross-validation, the training dataset is divided into k folds for each training compound, and k-1 (where k is any natural number, k>1) folds are used as first training data and one fold is used as first test data.

[0061] Note that the specific aspects of the method of this embodiment have been described above in the control process, so a detailed description thereof will be omitted here.

[0062] 3. Program In the program of this embodiment, an information processing device is caused to execute the steps of acquiring a training dataset including information on training compounds, information on curing conditions, and information on reaction rates, and constructing a training model based on the training dataset. The step of constructing the training model involves constructing the training model through performance evaluation of candidate models by double cross-validation, and in cross-validation outside the double cross-validation, the training dataset is divided into k folds for each training compound, and k-1 (where k is any natural number, k>1) folds are used as first training data and one fold is used as first test data.

[0063] The program may be recorded on a readable recording medium. Note that the specific aspects of the processing executed by the program of this embodiment have been described in the control processing section above, and therefore will not be described in detail here.

[0064] 4. Composition The composition of this embodiment includes a compound selected based on information about the reaction rates of the predicted compounds output by the information processing device or the information processing method. Specifically, a user may obtain information about the reaction rates of one or more predicted compounds based on information about one or more predicted compounds and information about curing conditions using the information processing device or the information processing method, select a predicted compound having a desired reaction rate from the information, and use the selected compound as a component of the composition. Alternatively, for example, the information processing device may repeat the above-described steps S01 and S02 while changing the information about the predicted compounds, and when the reaction rate of the predicted compound is equal to or greater than the user's desired reaction rate, select the predicted compound and use it as a component of the composition.

[0065] 5. Method for producing the composition The method for producing a composition according to this embodiment includes blending a compound selected based on information about the reaction rates of the predicted compounds output by the information processing device or the information processing method into the composition. Specifically, a user may obtain information about the reaction rates of one or more predicted compounds based on information about one or more predicted compounds and information about curing conditions using the information processing device or the information processing method, select a predicted compound having a desired reaction rate from the information, and blend the compound into the composition to produce the composition. Alternatively, for example, the information processing device may repeat steps S01 and S02 described above while changing the information about the predicted compounds, and when the reaction rate of the predicted compound becomes equal to or greater than the user's desired reaction rate, select the predicted compound and blend it into the composition to produce the composition. [Example]

[0066] The present invention will be described in more detail below using examples.

[0067] Compositions containing acrylate compounds A to H as monomers and an acylphosphine oxide photopolymerization initiator as a photopolymerization initiator were prepared, and then coated on a substrate and dried to form a coating film of the composition. Note that the mass ratio of the monomer to the photopolymerization initiator was the same in each composition.

[0068] The resulting coatings were then irradiated with UV light at seven different light intensities in the atmosphere to obtain cured films of each coating. IR measurements were performed on the resulting cured films, and the reaction rate was calculated from the intensity of the peak derived from the acrylate group. The relationship between the reaction rate and the light irradiation intensity for each coating is shown in Figure 2B.

[0069] The appropriate preprocessing method and regression analysis method were determined by double cross-validation using the dataset in Figure 2B. In the outer cross-validation, the dataset was divided into eight folds for each compound, with seven folds used as training data and one fold used as test data. In the inner cross-validation, three folds were used. The method with the highest coefficient of determination (r) was selected from all preprocessing methods and regression analysis methods.2 A learning model was constructed by using the combination with the highest value as the optimal preprocessing method and regression analysis method.

[0070] Using the created learning model, we predicted the reaction rates of predicted compounds other than compounds A to H. Figure 4A shows the predicted values ​​for the relationship between the reaction rate of the predicted compounds and the light irradiation intensity. Compositions were created using the two predicted compounds with high reaction rates, Proposal 1 and Proposal 2, respectively, and the reaction rates were measured using the same method as above. Note that the type of photopolymerization initiator used and the mass ratio of monomer to photopolymerization initiator in each composition using Proposal 1 and Proposal 2 were the same as those in the compositions using compounds A to H. As shown in Figure 4B, all of the proposed compounds exhibited reaction rates equivalent to or higher than that of the conventional product (compound C).

[0071] From the above examples, it was found that the information processing device etc. of this embodiment can construct an appropriate reaction rate prediction model even if the number of data used to create the learning model is small, and it is also possible to propose compounds with high reaction rates. [Explanation of symbols]

[0072] 1...information processing system, 10...coating film, 11...monomer, 12...photopolymerization initiator, 100...information processing device, 110...processor, 111...first acquisition unit, 112...learning unit, 113...second acquisition unit, 114...output unit, 120...communication interface, 130...input / output interface, 140...memory, 150...storage, 160...communication bus, 200...user device,

Claims

1. a first acquisition unit that acquires a training dataset including information on the training compound, information on the curing conditions, and information on the reaction rate; a learning unit that constructs a learning model based on the learning dataset, the learning unit constructs the learning model through performance evaluation of candidate models by double cross validation, In the cross validation outside the double cross validation, the training dataset is divided into k folds for each training compound, and k-1 folds (where k is any natural number, k>1) are used as first training data, and one fold is used as first test data. Information processing device.

2. a second acquisition unit that acquires information about the predicted compound; and an output unit that outputs information about the predicted compound and information about the reaction rate of the predicted compound based on the learning model. The information processing device according to claim 1 .

3. The information about the training compound includes information about atom positions and bond types. The information processing device according to claim 1 .

4. The training compound is a monomer and / or a photopolymerization initiator; The information processing device according to claim 1 .

5. the predicted compound includes a plurality of monomers, the output unit outputs information regarding the reaction rate of the predicted compound by a weighted average obtained by multiplying the reaction rate of each monomer by its content ratio. The information processing device according to claim 2 .

6. The information about the curing conditions includes information about the irradiation atmosphere. The information processing device according to claim 1 .

7. The information processing device obtaining a training data set including information on the training compound, information on the curing conditions, and information on the reaction rate; constructing a learning model based on the learning dataset, wherein the step of constructing the learning model involves constructing the learning model through performance evaluation of a candidate model by double cross-validation; In the cross validation outside the double cross validation, the training dataset is divided into k folds for each training compound, and k-1 folds (where k is any natural number, k>1) are used as first training data, and one fold is used as first test data. Information processing methods.

8. obtaining information about the predicted compound; and outputting information about the predicted compound and information about the reaction rate of the predicted compound based on the learning model. The information processing method according to claim 7.

9. In the information processing device, obtaining a training data set including information on the training compound, information on the curing conditions, and information on the reaction rate; constructing a learning model based on the training dataset; The step of constructing the learning model constructs the learning model through performance evaluation of a candidate model by double cross validation; In the cross validation outside the double cross validation, the training dataset is divided into k folds for each training compound, and k-1 folds (where k is any natural number, k>1) are used as first training data, and one fold is used as first test data. program.

10. obtaining information about the predicted compound; and outputting information about the predicted compound and information about the reaction rate of the predicted compound based on the learning model. The program according to claim 9.

11. The information processing device according to any one of claims 1 to 6 or the information processing method according to claim 7 or 8 includes a compound selected based on information about reaction rates output. composition.

12. a step of blending a compound selected based on the information on the reaction rate output by the information processing device according to any one of claims 1 to 6 or the information processing method according to claim 7 or 8 into a composition; Method for producing the composition.

Images