Prediction system, program, and prediction method

Abstract

Provided are a prediction system, a program, and a prediction method for generating a prediction model for predicting the performance of a composition. **Solution**: A prediction system for generating a prediction model for predicting the performance of a composition, comprising: a model acquisition unit that acquires, from a database, first composition group data including a plurality of pairs of composition information and performance values of the composition and is learned by a learning unit to obtain a first prediction model for predicting a performance value from the composition information; a data acquisition unit that acquires, from the database, second composition group data including a plurality of pairs of composition information and performance values of a composition different from the first composition group data; and a calculation unit that calculates parameters of a second prediction model based on the first prediction model and the second composition group data.

Description

Technical Field

[0001] The present invention relates to a prediction system, a program, and a prediction method.

Background Art

[0002] Patent Document 1 describes "an apparatus including: a composition acquisition unit that acquires composition data indicating the composition of a photosensitive resin composition; a property acquisition unit that acquires property data indicating the properties of the photosensitive resin composition; and a learning processing unit that executes a learning process of a model that outputs recommended composition data indicating the composition of a photosensitive resin composition recommended in response to input of target property data indicating the properties of a target photosensitive resin composition, using the acquired composition data and the property data including the property data." (Claim 1). [Prior Art Documents] [Patent Documents] [Patent Document 1] Japanese Unexamined Patent Application Publication No. 2020-77346

Summary of the Invention

[0003] In a first aspect of the present invention, a prediction system that generates a prediction model for predicting the performance of a composition is provided. The prediction system includes a model acquisition unit, a data acquisition unit, and a calculation unit. The model acquisition unit may acquire a first prediction model that predicts a performance value from composition information, which is learned based on first composition group data including a plurality of pairs of composition information and performance values of a composition. The data acquisition unit may acquire second composition group data including a plurality of pairs of composition information and performance values of a composition different from the first composition group data. The calculation unit may calculate parameters of a second prediction model based on the first prediction model and the second composition group data.

[0004] In the above, the calculation unit may calculate parameters of the second prediction model by correcting the first prediction model based on the second composition group data.

[0005] In the above, the model acquisition unit may acquire the first parameters that constitute the first prediction model. The calculation unit may include a prior distribution generation unit and a statistical calculation unit. The prior distribution generation unit may generate a first probability distribution of the first parameters. The statistical calculation unit may calculate, based on the first probability distribution and the second composition group data, a statistical value of a second probability distribution, which is a posterior probability distribution taking the first probability distribution as a prior probability and the second composition group data as observation data, as a parameter of the second prediction model.

[0006] In the above, the prior distribution generation unit may change the degree to which the first probability distribution is transferred to the second probability distribution by changing the dispersion degree of the first probability distribution.

[0007] In the above, the prior distribution generation unit may change the degree to which the first probability distribution is transferred to the second probability distribution so as to improve the accuracy of the second prediction model by performing cross-validation using the second composition group data.

[0008] In the above, the statistical value of the second probability distribution may be the average value of the second probability distribution.

[0009] In the above, the statistical calculation unit may calculate the average value of the second probability distribution by performing sampling.

[0010] In the above, the statistical calculation unit may perform sampling by the Markov chain Monte Carlo method (MCMC).

[0011] In the above, the data acquisition unit may acquire third composition group data including a plurality of pairs of composition information and performance values of compositions different from the first composition group data and the second composition group data. The statistical calculation unit may calculate, taking the second probability distribution as a new prior distribution, a statistical value of a third probability distribution, which is a posterior probability distribution taking the third composition group data as observation data, as a parameter of a third prediction model obtained by updating the second prediction model.

[0012] In the above, the first prediction model and the second prediction model may be regression models. The parameter may be a regression coefficient.

[0013] In the above, the number of pairs included in the first composition group data may be three times or more the number of pairs included in the second composition group data.

[0014] In the above, the data acquisition unit may acquire target composition data that is composition information of a target composition to be predicted. The prediction system may include a prediction unit that inputs the composition information of the target composition into a second prediction model to predict the performance value of the target composition.

[0015] In the above, the performance value may be within the range of products that can be manufactured by the composition.

[0016] In the above, the performance value may be the minimum dimension of a product that can be manufactured by the composition.

[0017] In the above, the composition may be a resin composition. The composition information may include information on polymers, monomers, and other components in the resin composition.

[0018] In a second aspect of the present invention, a prediction method including a model acquisition stage, a data acquisition stage, and a calculation stage is provided. In the model acquisition stage, based on first composition group data including a plurality of pairs of composition information and performance values of a composition, a first prediction model that regresses the performance value from the composition information may be acquired. In the data acquisition stage, second composition group data including a plurality of pairs of composition information and performance values of a composition different from the first composition group data may be acquired. In the calculation stage, based on the first prediction model and the second composition group data, parameters of the second prediction model may be calculated.

[0019] In a third aspect of the present invention, a program executed by a computer to cause the computer to function as the above prediction system is provided.

[0020] Note that the above summary of the invention does not list all the features of the present invention. Also, sub-combinations of these feature groups may also be inventions.

Brief Description of the Drawings

[0021]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Modes for Carrying Out the Invention

[0022] Hereinafter, the present invention will be described through embodiments of the invention. However, the following embodiments do not limit the invention according to the claims. Also, not all combinations of features described in the embodiments are essential for the solution means of the invention.

[0023] FIG. 1 shows the configuration of the prediction system 100 according to this embodiment. The prediction system 100 generates a prediction model for predicting the performance of a composition. The prediction system 100 predicts the performance value of a target composition using the prediction model. The prediction system 100 searches for a composition that satisfies a target performance value using the prediction model.

[0024] The composition is not particularly limited and may be, for example, a resin composition, such as a photosensitive resin composition. The photosensitive resin composition may be a negative type with photocurability or a positive type with photosolubility. The photosensitive resin composition may be in a liquid state, or may be in a state of being temporarily cured and / or dried. The temporarily cured and / or dried photosensitive resin composition may be in the form of a film formed on a support layer. In this case, it is also referred to as a photosensitive resin laminate.

[0025] In this embodiment, as an example, the photosensitive resin composition may be formed in a roll shape by being laminated with a support layer and a protective layer and wound up for the purpose of facilitating storage and transportation. In this case, the roll-shaped laminate is also referred to as a photosensitive resin wound body.

[0026] The prediction system 100 includes a learning unit 120, a model acquisition unit 130, a data acquisition unit 140, a calculation unit 150, a prediction unit 160, and a search unit 180. The prediction system 100 may include modules having other functions as necessary.

[0027] The prediction system 100 may be a computer such as a PC (personal computer), a tablet computer, a smartphone, a workstation, a server computer, or a general-purpose computer, or may be a computer system to which a plurality of computers are connected.

[0028] Alternatively, the prediction system 100 may be a dedicated computer designed for uses such as prediction model generation, or may be dedicated hardware realized by a dedicated circuit. The prediction system 100 may be implemented by a single device (computer), or may be realized by a plurality of devices with role sharing. In the prediction system 100, although not particularly described below, a memory / hard disk, etc. are provided, information necessary for processing is appropriately stored, and information is transmitted between each processing module such as the learning unit 120 and the model acquisition unit 130.

[0029] The learning unit 120 learns a first prediction model that predicts performance values from composition information. The learning unit 120 acquires first composition group data including a plurality of pairs of composition information and performance values of compositions from the database 110.

[0030] Based on the first composition group data, the learning unit 120 may learn the first prediction model. The learning unit 120 may acquire, by learning, first parameters that constitute the first prediction model.

[0031] The model acquisition unit 130 acquires the first prediction model from the learning unit 120. The prediction system 100 may not have the learning unit 120. In that case, the model acquisition unit 130 may acquire a first prediction model that has been previously learned or generated from the database 110 or externally. The model acquisition unit 130 may acquire the first parameters from the learning unit 120.

[0032] The data acquisition unit 140 may acquire data necessary for the prediction system 100 from the database 110 and provide it to each module of the prediction system 100. For example, the data acquisition unit 140 may acquire second composition group data including a plurality of pairs of composition information and performance values of compositions different from the first composition group data.

[0033] The second composition group data is data on a smaller scale than the first composition group data. That is, the number of pairs included in the second composition group data is smaller than the number of pairs included in the first composition group data. The second composition group data may be data of the same scale as or larger than the first composition group data.

[0034] The data acquisition unit 140 may further acquire third composition group data including a plurality of pairs of composition information and performance values of compositions different from the first composition group data and the second composition group data. The data acquisition unit 140 may acquire target composition data that is the composition information of the target composition to be predicted.

[0035] The calculation unit 150 generates a second prediction model. The calculation unit 150 may obtain the first prediction model from the model acquisition unit 130 and obtain the first composition group data from the data acquisition unit 140. The calculation unit 150 may calculate the parameters of the second prediction model based on the first prediction model and the second composition group data. The calculation unit 150 may calculate the parameters of the second prediction model by correcting the first prediction model based on the second composition group data.

[0036] FIG. 2 shows an example of the configuration of the calculation unit 150 according to the present embodiment. The calculation unit 150 may include a prior distribution generation unit 152 and a statistical calculation unit 154.

[0037] The prior distribution generation unit 152 generates a first probability distribution of the first parameters of the first prediction model. The prior distribution generation unit 152 supplies the first probability distribution to the statistical calculation unit 154.

[0038] Based on the first probability distribution and the second composition group data, the statistical calculation unit 154 calculates the statistical value of the second probability distribution, which is the posterior probability distribution with the first probability distribution as the prior probability and the second composition group data as the observation data, as the second parameters of the second prediction model. That is, the statistical calculation unit 154 generates the second prediction model by updating the first prediction model represented by the first probability distribution with the second composition group data, which is the observation data. Details of the operations of the prior distribution generation unit 152 and the statistical calculation unit 154 will be described later.

[0039] The prediction unit 160 predicts the performance value of the composition based on the second prediction model. For example, the prediction unit 160 inputs the composition information (hereinafter also referred to as "target composition information") of the composition to be predicted (hereinafter also referred to as "target composition") into the second prediction model and predicts the performance value of the target composition. The prediction unit 160 obtains the performance value output from the second prediction model as the prediction result of the performance value of the target composition.

[0040] The exploration unit 170 explores a composition having a performance value that satisfies the target performance value. The exploration unit 170 receives an input of the target performance value, supplies target composition information to the prediction unit 160, and acquires the predicted performance value from the prediction unit 160. The exploration unit 170 outputs, as recommended composition information, the target composition information such that the predicted performance value satisfies the target performance value.

[0041] FIG. 3 shows an example of the manufacturing apparatus 200 according to the present embodiment. The manufacturing apparatus 200 manufactures a composition. The manufacturing apparatus 200 manufactures, for example, a photosensitive resin composition. The manufacturing apparatus 200 may include a mixing unit 210 that mixes raw materials of the composition and a generation unit 220 that generates a composition from the mixed raw materials.

[0042] The mixing unit 210 may perform mixing based on the recommended composition information output from the exploration unit 170. For example, the mixing unit 210 may receive the recommended composition information from the prediction system 100 and mix the raw materials according to the recommended composition information.

[0043] The generation unit 220 may include a filtering unit that filters the raw materials, a coating unit that coats the raw materials on a support layer to form the composition in a film shape, a roller unit that laminates a protective layer on the composition and winds up the laminate, and the like.

[0044] As described above, according to the present embodiment, the second prediction model is generated by correcting the first prediction model based on the first composition group data based on the second composition group data. Even when the scale of the second composition group data is small, the second prediction model can be generated with high accuracy.

[0045] Thereby, the accuracy of the second prediction model can be improved as compared with the case where the second composition group data is used alone. Further, as compared with the case of re-learning using both the first composition group data and the second composition group data, the accuracy of the prediction model can be further improved, and the calculation resources and storage resources to be used can be reduced. This is because a model based on the second composition group data can be constructed while adjusting the weight of the first composition group data.

[0046] Figure 4 shows the flow of the prediction method according to this embodiment. The prediction system 100 manufactures a composition of recommended composition information that satisfies the target performance value by using the second prediction model, for example, by executing each process of S10 to S50. The order of the processes of S10 to S50 may be changed, and some processes may be omitted.

[0047] In S10, the prediction system 100 generates a second prediction model.

[0048] Figure 5 shows an example of the sub-flow of S10 in Figure 4. The prediction system 100 may perform S10 by executing the flow of S100 to S400 in Figure 5. S100 to S400 may be in a different order, and some may be omitted.

[0049] First, in S100, the learning unit 120 acquires first composition group data. The first composition group data includes a plurality of pairs of composition information of the composition and performance values.

[0050] The composition information is information regarding components included in the composition and / or raw materials that can generate the composition (hereinafter, components and raw materials are also collectively referred to as "component equivalents").

[0051] The composition information may be, for example, the presence or absence of component equivalents in the composition, the content of component equivalents, and / or the content ratio of component equivalents. The component equivalents may be represented by information that can at least partially specify the chemical structure, and may be represented by, for example, a common name, a conventional name, an IUPAC name, a trade name, a model number, a chemical formula, a structural formula, a composition formula, a CAS number, a three-dimensional structure, a molecular fingerprint, a molecular weight, an average molecular weight, a molecular weight distribution, a manufacturer, a manufacturing location, and / or a manufacturing lot.

[0052] The composition may be, for example, a resin composition, and as an example, a photosensitive resin composition. However, the composition is not limited thereto. The composition information may include information on polymers, monomers, and other components in the resin composition as component equivalents.

[0053] More specifically, the components, etc. may include at least one of an alkali-soluble polymer, an ethylenically unsaturated bond-containing compound, a photopolymerization initiator, a resin having a repeating unit containing an acid-decomposable group (a group that is deprotected by an acid as an example), a phenolic resin, a photoacid generator, a dissolution inhibitor, a sensitizer, a polymerization inhibitor, an adhesion promoter, a plasticizer, and a solvent.

[0054] Among these, the alkali-soluble polymer may be a polymer in the photocurable resin composition, for example, a polymer containing a carboxyl group. The ethylenically unsaturated bond-containing compound may be a monomer in the photocurable resin composition. The photopolymerization initiator may bond monomers to each other by exposure in the photocurable resin composition. The resin having a repeating unit containing an acid-decomposable group and the phenolic resin may be polymers in the photosoluble resin composition and may be dissolved and decomposed by an acid. The photoacid generator generates an acid by exposure in the photosoluble resin composition. The dissolution inhibitor, also referred to as a dissolution suppressing agent, suppresses the dissolution of the components in an aqueous alkali solution. The sensitizer may be a photosensitizer as an example, but may also be other types of sensitizers such as N-phenylglycine. The polymerization inhibitor may inhibit the polymerization reaction due to the influence of light or heat. The adhesion promoter enhances the adhesion of the photosensitive resin composition to the substrate surface. The plasticizer is added to impart flexibility to the photosensitive resin composition or to facilitate processing.

[0055] The performance value may be a numerical representation of the performance of the composition itself, the performance after the composition is processed by a predetermined method, or the performance of the product manufactured from the composition. For example, the performance value may be one or more of the minimum development time of the photosensitive resin composition, sensitivity to light, transparency, resolution, the range of products manufacturable from the composition, adhesion to the substrate, developing solution foaming property, developing solution aggregating property, edge haze characteristics, cured film flexibility, hue stability, peeling time, peeling piece size, film thickness, tackiness to the support layer or protective layer, and tenting property.

[0056] The development of a photosensitive resin composition may involve exposing the photosensitive resin composition, curing or solubilizing the photosensitive resin composition in the exposed area, and then removing the photosensitive resin composition in the exposed area or the unexposed area to reveal a negative or positive image corresponding to the exposed area. When the photosensitive resin composition is photocurable (also referred to as negative type), the minimum development time, also known as the breakpoint, indicates the minimum time required for the photosensitive resin composition to be developed. As an example, the minimum development time may be the time it takes for all of the photosensitive resin composition laminated on a substrate to be removed by spraying an alkaline solution without exposing the photosensitive resin composition. The minimum development time may be the time when parameters that can affect the development time, such as the number of spray nozzles and the spraying pressure, are set as fixed values.

[0057] When the photosensitive resin composition is photocurable, the sensitivity to light, also known as the minimum curing exposure amount, indicates the minimum exposure amount capable of forming an image corresponding to the exposed area. For example, when the photosensitive resin composition is photocurable, the sensitivity to light may be the minimum exposure amount (mJ / cm 2 ) at which the photosensitive resin composition remaining and cured on the substrate occurs when the photosensitive resin composition applied or laminated on the substrate is exposed and developed. As an example, the sensitivity to light may be calculated from the lowest transmittance at which the photosensitive resin composition is cured by exposing the photosensitive resin composition using a mask with different transmittances step by step.

[0058] Transparency, also referred to as transmittance, indicates the light transmittance. The wavelength of the light transmitted may be the wavelength that cures or solubilizes the photosensitive resin composition.

[0059] The range of products that can be manufactured using the composition may be dimensions and the like that can be formed without defects using the photosensitive resin composition. For example, the range of products that can be manufactured using the composition may be the minimum dimensions of the products that can be manufactured using the composition.

[0060] The developability of the developer indicates the foaming property when developing the photosensitive resin composition using the developer. As the developability of the developer, values measured by various known methods may be used.

[0061] The aggregability of the developer indicates the aggregating property when developing the photosensitive resin composition using the developer. As the aggregability of the developer, values measured by various known methods may be used.

[0062] The edge squeeze-out property indicates the amount of the photosensitive resin composition that protrudes outward from the end face of the photosensitive resin laminate roll due to the winding pressure during storage of the photosensitive resin laminate roll. The smaller the edge squeeze-out property, the longer the shelf life of the photosensitive resin laminate, which is preferable. As the edge squeeze-out property, values measured by various known methods may be used.

[0063] The flexibility of the cured film indicates the flexibility of the developed photocurable photosensitive resin composition. For example, the flexibility of the cured film may be the minimum diameter at which no cracks occur in the resist when the photosensitive resin laminate of the photosensitive resin composition laminated on a flexible substrate is exposed and developed and then wound around a plurality of cylinders with different diameters. As an example, the flexibility of the cured film may be measured by a mandrel bending tester.

[0064] The tackiness with the support layer or the protective layer indicates the adhesiveness between the photosensitive resin laminate and the support layer or the protective layer. For example, the tackiness may be the force required for peeling when the photosensitive resin laminate is peeled from the support layer or the protective layer using a tensilon device. Since the usability of the photosensitive resin laminate roll deteriorates both when the tackiness is excessive and when it is too small, it is preferably within an appropriate range.

[0065] The peeling time indicates the peelability when peeling the photosensitive resin composition from the substrate. For example, the peeling time may be the time until the photosensitive resin composition peels from the substrate when the photosensitive resin laminate laminated and exposed on the substrate is immersed in an alkaline peeling solution for development. A shorter peeling time is preferable.

[0066] The hue stability indicates the stability of the color of the photosensitive resin composition. As the hue stability, values measured by various known methods may be used.

[0067] The peel piece size indicates the size of the photosensitive resin composition peeled from the substrate (also referred to as a peel piece). The peel piece size may be the size of the peel piece peeled from the substrate and fragmented by a water washing spray. It is preferable that the peel piece is smaller.

[0068] The tenting property indicates the breakage rate when tenting the tent holes of the substrate with the photosensitive resin laminate. For example, the tenting property may be the number of broken tent holes when laminating, exposing, and developing a photosensitive resin laminate on a substrate having tent holes with a diameter of 1 to 10 mm. The smaller the tenting property, the more preferable.

[0069] Next, the learning unit 120 learns the first prediction model. By learning the first prediction model, the learning unit 120 obtains the first parameters of the first prediction model that constitutes the prediction model. The first prediction model may be one that inputs composition information and outputs a performance value.

[0070] For example, the learning unit 120 may learn a regression model as the first prediction model. In this case, the first parameters of the first prediction model may include regression coefficients and / or intercepts. The regression model is not particularly limited and may be, for example, a linear regression model such as simple regression, multiple regression, logistic regression, or a non-linear regression model. The learning unit 120 may perform learning by a known algorithm.

[0071] The learning unit 120 may learn other than a regression model as the first prediction model. For example, the learning unit 120 may learn a neural network such as a recurrent type or a time delay type, random forest, gradient boosting, logistic regression, or a support vector machine (SVM). The learning unit 120 may obtain the parameters (for example, the weights of each node of the neural network) that constitute these prediction models as the first parameters of the first prediction model.

[0072] Instead of the learning unit 120 acquiring the first composition group data, the data acquisition unit 140 may acquire the first composition group data and provide it to the learning unit 120.

[0073] FIG. 6 shows an example of composition group data. The learning unit 120 may learn the first prediction model using the data as shown in FIG. 6 as the first composition group data.

[0074] In FIG. 6, "Polymer 1", "Polymer 2", "Monomer 1", and "Monomer 2" correspond to components of the composition information and the like. Polymer 1 and Polymer 2 represent specific types of polymers. Monomer 1 and Monomer 2 represent specific types of monomers. For example, in FIG. 6, the composition of ID = 1 contains 0.5 parts by weight of a polymer of the type "Polymer 1", 0.1 parts by weight of a polymer of the type "Polymer 2", 0.4 parts by weight of a monomer of the type "Monomer 1", and 0.1 parts by weight of a monomer of the type "Monomer 2", and as a result, it is shown that the performance value was 4.2.

[0075] Next, in S300, the data acquisition unit 140 acquires the second composition group data. The second composition group data may be in the same format as the first composition group data (for example, the same format as shown in FIG. 6). The number of pairs included in the second composition group data may be less than the number of pairs included in the first composition group data. For example, the number of pairs included in the first composition group data may be 3 times or more, 5 times or more, or 10 times or more the number of pairs included in the second composition group data.

[0076] The composition information of the second composition group data may be included in the range of the composition information of the first composition group data. For example, there may be no components etc. included in the composition information of the second composition group data that are not present in the components etc. included in the composition information of the first composition group data (that is, the second composition group data includes only the components etc. that have already appeared).

[0077] Alternatively, the composition information of the second composition group data may include those outside the range of the composition information of the first composition group data. That is, the second composition group data may include new components, etc. (hereinafter referred to as "new components, etc.") that are not included in the components, etc. (hereinafter referred to as "existing components, etc.") included in the first composition group data.

[0078] In this case, the data acquisition unit 140 may replace the information of the existing components, etc. based on a predetermined rule with the information of the new components, etc. For example, the content of the new components, etc. may be replaced with the content of the existing components, etc. having similar chemical properties. The content may be the same before and after replacement, or the content may be corrected based on a predetermined rule. The replacement rule may be registered in advance by the user, or the user may execute the replacement on the spot. For the new components, etc., in addition to / instead of the above, application or suppression of transfer of the no-information distribution described later may be performed.

[0079] As an example, when methyl methacrylate is included as a component, etc. in the first composition group data but methyl acrylate is not included, and methyl methacrylate appears as a component, etc. in the second composition group data, the data acquisition unit 140 may replace the content of methyl methacrylate (for example, 0.2 parts by weight of methyl methacrylate) in the second composition group data with the same content of methyl acrylate (for example, 0.2 parts by weight of methyl acrylate).

[0080] The performance value of the second composition group data may be obtained by the same method as the performance value of the first composition group data. For example, the performance value of the first composition group data may represent the performance after processing the composition under predetermined conditions (for example, exposure, development, and heating under certain conditions), and the performance value of the second composition group data may represent the performance after processing the composition under the same conditions.

[0081] Alternatively, the performance values of the second composition group data may be obtained by a method different from that of the first composition group data. For example, the performance values of the first composition group data may represent the performance after processing the composition under predetermined conditions (e.g., exposure, development, and heating under certain conditions), and the performance values of the second composition group data may represent the performance after processing the composition under different conditions.

[0082] In the second composition group data, the conditions of the processing performed on the composition in the first composition group data (e.g., temperature, processing time, atmospheric pressure, pressure, and / or voltage, etc.) may be changed. In the second composition group data, a part of the processing performed on the composition in the first composition group data may be omitted. In the second composition group data, in addition to the processing performed on the composition in the first composition group data, another additional processing may be executed. As an example, the performance value of the first composition group data may be the performance value when heating is performed on the composition after exposure and before development, and the performance value of the second composition group data may be the performance value when heating after exposure is not performed.

[0083] Next, in S400, the calculation unit 150 generates a second prediction model.

[0084] FIG. 7 shows an example of the sub-flow of S400 in FIG. 5. The calculation unit 150 may perform S400 by executing the flows of S410 to S440 in FIG. 7.

[0085] First, in S410, the prior distribution generation unit 152 generates a first probability distribution of the first parameters of the first prediction model learned in S200. For example, the prior distribution generation unit 152 generates the first probability distribution by setting the dispersion degree specified by the user or predetermined for the distribution having the regression coefficient of the first prediction model as the average value. The prior distribution generation unit 152 may generate a distribution having a predetermined shape (e.g., Gaussian distribution) as the first probability distribution.

[0086] The prior distribution generation unit 152 changes the degree to which the first probability distribution is later transferred to the second probability distribution by changing the degree of dispersion of the first probability distribution. When the degree of dispersion increases, the first probability distribution becomes closer to a uniform distribution and the degree of transfer decreases. When the degree of dispersion decreases, the peak of the first probability distribution becomes sharper and the degree of transfer increases. The degree of dispersion may be, for example, the variance or standard deviation of the distribution. For new components or the like, the prior distribution generation unit 152 may generate an uninformative distribution (i.e., a uniform distribution) as the first probability distribution, or may not execute the transfer.

[0087] The prior distribution generation unit 152 may make the degrees of transfer of a plurality of parameters the same. For example, the prior distribution generation unit 152 may make the degree of transfer (e.g., degree of dispersion) for the parameters (e.g., regression coefficients) related to Polymer 1 the same as the degree of transfer (e.g., degree of dispersion) for the parameters (e.g., regression coefficients) related to Polymer 2.

[0088] The prior distribution generation unit 152 may make the degrees of transfer of a plurality of parameters different. For example, the second composition group data may have different degrees of transfer (e.g., degrees of dispersion) for the parameters (e.g., regression coefficients) related to Polymer 1 and the parameters (e.g., regression coefficients) related to Polymer 2.

[0089] In the first S410, the prior distribution generation unit 152 may randomly determine the degree of transfer, or may input the degree of transfer from the user.

[0090] Next, in S420, the statistical calculation unit 154 calculates the second parameters of the second prediction model. The statistical calculation unit 154 may calculate the statistical values of the second probability distribution, which is the posterior probability distribution with the first probability distribution as the prior probability and the second composition group data as the observed data, as the second parameters of the second prediction model based on the first probability distribution generated in S410 and the second composition group data acquired in S300.

[0091] The statistical calculation unit 154 may calculate the statistical value of the second probability distribution by performing sampling on the parameters of the prediction model. For example, the statistical calculation unit 154 may perform sampling by the Markov chain Monte Carlo method (MCMC). The statistical value may be, for example, an average value, a median value, or an expected value. For example, the statistical calculation unit 154 acquires the average value of the second probability distribution obtained as a result of sampling as the second parameter of the second prediction model.

[0092] The second prediction model may be a model similar to the first prediction model. For example, when the first prediction model is a regression model, the second prediction model may be a regression model of the same type as the first prediction model. In this case, the second parameter may be a regression coefficient and / or an intercept. In this way, the statistical calculation unit 154 uses the first prediction model as a prior distribution, corrects the prior distribution using the second composition group data as observations to obtain a posterior distribution, and analyzes the posterior distribution to obtain the second parameter.

[0093] Next, in S430, the prior distribution generation unit 152 acquires the prediction accuracy of the second prediction model. The prior distribution generation unit 152 may obtain the prediction accuracy by cross-validation.

[0094] For example, the prior distribution generation unit 152 may input the composition information of the second composition group data into the second prediction model, and compare the output value with the performance value of the second composition group data to evaluate the accuracy of the current second prediction model. As an example, the prior distribution generation unit 152 may calculate the degree of divergence between the output value and the performance value by a known loss function, and calculate the accuracy of the second prediction model based on the degree of divergence.

[0095] Next, in S440, the prior distribution generation unit 152 determines whether to end the process of S400. The prior distribution generation unit 152 may end the process of S400 when the accuracy of the second prediction model is equal to or higher than a threshold value. The prior distribution generation unit 152 may end the process of S400 when the loop of S410 to S440 is executed a predetermined number of times and / or for a predetermined time.

[0096] The prior distribution generation unit 152 may output the last generated second parameter and / or the most accurate second parameter as the second parameter of the final second prediction model. If the end condition is not satisfied, the prior distribution generation unit 152 returns the process to S410 to generate the first probability distribution again.

[0097] In the second and subsequent executions of S410, the prior distribution generation unit 152 may lower or increase the degree of transition (e.g., the degree of distribution). For example, the prior distribution generation unit 152 may increase the degree of dispersion while maintaining the average value of the first probability distribution. The prior distribution generation unit 152 may adjust the degree of transition so as to increase the accuracy based on the accuracy evaluated in S430. For example, the prior distribution generation unit 152 may adjust the degree of transition by a gradient method or the like.

[0098] Thus, in S400, the prior distribution generation unit 152 may change the degree to which the first probability distribution is transferred to the second probability distribution so as to increase the accuracy of the second prediction model by performing cross-validation using the second composition group data.

[0099] The processes by the prior distribution generation unit 152 and the statistical calculation unit 154 may be executed by known software. For example, programs such as pyMC, stan, and pyro may be used.

[0100] Thus, in S10, the calculation unit 150 generates a second prediction model that predicts the performance value from the composition information. The calculation unit 150 supplies the second prediction model to the prediction unit 160. The prediction unit 160 can input the target composition information of the target composition to the second prediction model and obtain the predicted value of the performance value of the target composition.

[0101] In S20, the search unit 170 inputs the target performance value from the user. The target performance value is the performance value that the composition should satisfy.

[0102] In S30, the search unit 170 searches for composition information of a composition that satisfies the target performance value. The search unit 170 generates target composition information for search and supplies this to the prediction unit 160. The prediction unit 160 inputs the target composition information into the second prediction model and obtains the performance value corresponding to the target composition information.

[0103] The search unit 170 may generate composition information so that the composition satisfies the target performance value. The search unit 170 may generate composition information by a known search algorithm. For example, initially, the search unit 170 may generate composition information randomly, and then generate composition information so that the performance value approaches the target performance value by a gradient method or the like.

[0104] The search unit 170 specifies the composition information that satisfies the target performance value as the recommended composition information. When the composition information that satisfies the target performance value cannot be found, the search unit 170 specifies one or a plurality of composition information that has obtained a performance close to the target performance value as the recommended composition information.

[0105] In S40, the search unit 170 outputs the generated recommended composition information. The search unit 170 may output the recommended composition information to the manufacturing apparatus 200.

[0106] In S50, the manufacturing apparatus 20 may manufacture a composition based on the recommended composition information. For example, the mixing unit 210 executes mixing of raw materials based on the recommended composition information. Thereby, the manufacturing apparatus 20 can manufacture a composition with a high possibility of achieving the target performance. After that, the performance value of the manufactured composition may be measured, and additional learning may be performed using the pair of the composition information and the performance value of such a composition.

[0107] The manufacturing apparatus 20 may manufacture a product using a composition. For example, the manufacturing apparatus 20 may apply a fluid of a composition, which is a photosensitive resin composition, onto a support layer. After applying the photosensitive resin composition, the manufacturing apparatus 20 may remove its solvent and dispersion medium by drying. Thereby, a laminate of the support layer and the film-shaped photosensitive resin composition is manufactured. The manufacturing apparatus 20 may provide a protective layer for protecting the photosensitive resin composition on the photosensitive resin composition to manufacture a photosensitive resin laminate.

[0108] According to the present embodiment, the prediction system 100 can generate a second prediction model based on relatively small-scale second composition group data by utilizing the first prediction model. According to such a method, calculation resources and / or storage resources can be saved as compared with the case of merging and re-learning the first composition group data and the second composition group data. Further, according to the present embodiment, even when the composition information, processing conditions, etc. of the second composition group data are different from those of the first composition group data, the second prediction model can be generated with high accuracy by referring to the first prediction model.

[0109] FIG. 8 shows another example of the sub-flow of S20 according to a modification of the present embodiment. In this modification, S20 may be executed by S100 to 600. S100 to S400 may be executed in the same manner as the processes described with reference to FIG. 5. In this modification, the prediction system 100 may execute S500 to S600 in addition to S100 to S400.

[0110] In S500, the data acquisition unit 140 acquires third composition group data. The third composition group data may be in the same format as the first composition group data and / or the second composition group data (for example, the same format as the format shown in FIG. 6). The number of pairs included in the third composition group data may be less than the number of pairs included in the first composition group data. For example, the number of pairs included in the first composition group data may be 3 times or more, 5 times or more, or 10 times or more the number of pairs included in the third composition group data.

[0111] The compositional information of the third group of composition data may be included within the range of the compositional information of the first group of composition data. Alternatively, the compositional information of the third group of composition data may include information outside the range of the compositional information of the first group of composition data. The data acquisition unit 140 may perform the same processing on the third group of composition data as that for the second group of composition data in S300.

[0112] In S600, the calculation unit 150 generates a third prediction model. The calculation unit 150 may generate the third prediction model in the same manner as the generation of the second prediction model in S400. For example, the statistical calculation unit 154 may use the second probability distribution obtained as the posterior probability distribution in S420 as a new prior distribution, and calculate the statistical value of the third probability distribution, which is the posterior probability distribution of observing the third group of composition data, as the parameters of the third prediction model obtained by updating the second prediction model. The calculation unit 150 supplies the third prediction model to the prediction unit 160, and the prediction unit 160 may use the third prediction model for prediction.

[0113] The prediction system 100 may execute the processing of S500 to S600 not at S10 but at any point after S20. For example, when the performance value of the composition is measured after manufacturing the composition in S50, the prediction system 100 may acquire the third group of composition data including the compositional information and performance value of the composition and generate a third prediction model.

[0114] The prediction system 100 may further acquire the composition data after the third group of composition data and further update the prediction model. For example, the prediction system 100 may acquire the fourth group of composition data in the same manner as S500 to S600 and generate a fourth prediction model.

[0115] Thus, according to this modification example, the calculation unit 150 can update the prediction model each time new composition data is acquired. Thereby, the prediction system 100 can improve the prediction accuracy using less computational resources and / or storage resources.

[0116] In the above example, a method of transferring the first prediction model to the second prediction model using the prior probability distribution in S400 was described, but the method is not limited thereto. For example, an error function between the first parameter of the first prediction model and the second parameter of the second prediction model may be defined, and the degree of transfer to the second prediction model may be adjusted by adjusting the magnitude of the error.

[0117] Various embodiments of the present invention may be described with reference to flowcharts and block diagrams, where the blocks may represent (1) stages of a process in which an operation is performed or (2) sections of an apparatus having a role of performing an operation. Specific stages and sections may be implemented by a dedicated circuit, a programmable circuit supplied with computer-readable instructions stored on a computer-readable medium, and / or a processor supplied with computer-readable instructions stored on a computer-readable medium. The dedicated circuit may include digital and / or analog hardware circuits, and may include an integrated circuit (IC) and / or discrete circuits. The programmable circuit may include a reconfigurable hardware circuit including memory elements such as logical AND, logical OR, logical XOR, logical NAND, logical NOR, and other logical operations, flip-flops, registers, a field programmable gate array (FPGA), a programmable logic array (PLA), etc.

[0118] A computer-readable medium may include any tangible device that can store instructions executable by an appropriate device. As a result, a computer-readable medium having instructions stored therein will comprise a product that includes instructions that can be executed to create means for performing the operations specified in a flowchart or block diagram. Examples of computer-readable media may include electronic memory media, magnetic memory media, optical memory media, electromagnetic memory media, semiconductor memory media, and the like. More specific examples of computer-readable media may include floppy (registered trademark) disks, diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disc read-only memory (CD-ROM), digital versatile disc (DVD), Blu-ray (registered trademark) disc, memory stick, integrated circuit card, and the like.

[0119] Computer-readable instructions may include any combination of one or more programming languages, including source code or object code written in any combination of assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or object-oriented programming languages such as Smalltalk (registered trademark), JAVA (registered trademark), C++, and conventional procedural programming languages such as the "C" programming language or similar programming languages.

[0120] Computer-readable instructions may be provided to a processor or programmable circuitry of a programmable data processing apparatus such as a general purpose computer, a special purpose computer, or other computers, either locally or via a wide area network (WAN) such as a local area network (LAN), the Internet, etc., and may execute the computer-readable instructions to create means for performing the operations specified in the flowchart or block diagram. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0121] FIG. 9 shows an example of a computer 2200 in which multiple aspects of the present invention may be embodied, in whole or in part. Programs installed on the computer 2200 may cause the computer 2200 to function as an operation associated with the apparatus according to an embodiment of the present invention or as one or more sections of the apparatus, or may cause the operation or the one or more sections to be executed, and / or may cause the computer 2200 to execute a process according to an embodiment of the present invention or a stage of the process. Such programs may be executed by the CPU 2212 to cause the computer 2200 to perform specific operations associated with some or all of the blocks of the flowcharts and block diagrams described herein.

[0122] The computer 2200 according to this embodiment includes a CPU 2212, a RAM 2214, a graphic controller 2216, and a display device 2218, which are interconnected by a host controller 2210. The computer 2200 also includes an input / output unit such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card drive, which are connected to the host controller 2210 via an input / output controller 2220. The computer also includes legacy input / output units such as a ROM 2230 and a keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

[0123] The CPU 2212 operates according to programs stored in the ROM 2230 and the RAM 2214, thereby controlling each unit. The graphic controller 2216 acquires image data generated by the CPU 2212 in a frame buffer or the like provided in the RAM 2214 or in itself, and causes the image data to be displayed on the display device 2218.

[0124] The communication interface 2222 communicates with other electronic devices via a network. The hard disk drive 2224 stores programs and data used by the CPU 2212 in the computer 2200. The DVD-ROM drive 2226 reads a program or data from the DVD-ROM 2201 and provides the program or data to the hard disk drive 2224 via the RAM 2214. The IC card drive reads programs and data from an IC card and / or writes programs and data to the IC card.

[0125] The ROM 2230 stores therein a boot program or the like executed by the computer 2200 upon activation, and / or a program dependent on the hardware of the computer 2200. The input / output chip 2240 may also be connected to the input / output controller 2220 via various input / output units such as a parallel port, a serial port, a keyboard port, a mouse port, etc.

[0126] The program is provided by a computer-readable medium such as a DVD-ROM 2201 or an IC card. The program is read from the computer-readable medium, installed in the hard disk drive 2224, the RAM 2214, or the ROM 2230, which are also examples of computer-readable media, and executed by the CPU 2212. The information processing described in these programs is read by the computer 2200, resulting in the cooperation between the programs and the various types of hardware resources described above. The apparatus or method may be configured by realizing the operation or processing of information according to the use of the computer 2200.

[0127] For example, when communication is executed between the computer 2200 and an external device, the CPU 2212 may execute a communication program loaded in the RAM 2214 and instruct the communication interface 2222 to perform communication processing based on the processing described in the communication program. The communication interface 2222 reads the transmission data stored in the transmission buffer processing area provided in a recording medium such as the RAM 2214, the hard disk drive 2224, the DVD-ROM 2201, or the IC card under the control of the CPU 2212, transmits the read transmission data to the network, or writes the received data received from the network to the reception buffer processing area or the like provided on the recording medium.

[0128] The CPU 2212 may cause all or a necessary part of a file or database stored in an external recording medium such as a hard disk drive 2224, a DVD-ROM drive 2226 (DVD-ROM 2201), an IC card, etc. to be read into the RAM 2214, and may execute various types of processing on the data on the RAM 2214. The CPU 2212 then writes back the processed data to the external recording medium.

[0129] Various types of information such as various types of programs, data, tables, and databases may be stored in the recording medium and may undergo information processing. The CPU 2212 may perform various types of processing on the data read from the RAM 2214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc. described throughout this disclosure and specified by the instruction sequence of the program, and write back the results to the RAM 2214. Also, the CPU 2212 may search for information in files, databases, etc. within the recording medium. For example, when a plurality of entries each having an attribute value of a first attribute associated with an attribute value of a second attribute are stored in the recording medium, the CPU 2212 searches for an entry that matches the condition where the attribute value of the first attribute is specified from among the plurality of entries, reads the attribute value of the second attribute stored in the entry, and thereby may obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0130] The programs or software modules described above may be stored in a computer-readable medium on or near the computer 2200. Also, a recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can be used as a computer-readable medium, thereby providing the program to the computer 2200 via the network.

[0131] As described above, the present invention has been described using embodiments, but the technical scope of the present invention is not limited to the scope described in the above embodiments. It is obvious to those skilled in the art that various changes or improvements can be made to the above embodiments. It is clear from the description of the claims that forms with such changes or improvements can also be included in the technical scope of the present invention.

[0132] In the claims, the specification, and the drawings, the execution order of each process such as operations, procedures, steps, and stages in the apparatus, system, program, and method shown is not explicitly stated as "earlier" or "preceding" etc., and it should be noted that it can be realized in any order unless the output of the previous process is used in the subsequent process. Regarding the operation flow in the claims, the specification, and the drawings, even if it is described using "first," "next," etc. for convenience, it does not mean that it is essential to implement in this order. The notation "A and / or B" may indicate "A, B, or A and C." The notation "A, B, and / or C" may indicate "any one of A, B, and C, or any combination of two or more of these."

Explanation of Reference Numerals

[0133] 100 Prediction System 110 Database 120 Learning Unit 130 Model Acquisition Unit 140 Data Acquisition Unit 150 Calculation Unit 152 Prior Distribution Generation Unit 154 Calculation Unit 160 Prediction Unit 170 Search Unit 200 Manufacturing Apparatus 210 Mixing Unit 220 Generation Unit 2200 Computer 2201 DVD-ROM 2210 Host Controller 2212 CPU 2214 RAM 2216 Graphic Controller 2218 Display Device 2220 Input / Output Controller 2222 Communication Interface 2224 Hard Disk Drive 2226 DVD-ROM Drive 2230 ROM 2240 Input / Output Chip 2242 Keyboard

Claims

1. 1. A prediction system for generating a predictive model for predicting the performance of a composition, comprising: a model acquisition unit that acquires a first prediction model that is trained based on first composition group data including a plurality of pairs of composition information and performance values ​​of compositions and predicts the performance value from the composition information; a data acquisition unit that acquires second composition group data including a plurality of pairs of composition information and performance values ​​of compositions different from the first composition group data; a calculation unit that calculates parameters of a second prediction model by correcting the first prediction model based on the first prediction model and the second composition group data; Equipped with The model acquisition unit acquires a first parameter constituting the first prediction model, The calculation unit is a prior distribution generating unit for generating a first probability distribution of the first parameter; a statistics calculation unit that calculates, based on the first probability distribution and the second composition group data, a statistical value of a second probability distribution, which is a posterior probability distribution having the first probability distribution as a priori probability and the second composition group data as observed data, as a second parameter of the second prediction model; having The prior distribution generating unit Varying the variance of the first probability distribution to change the degree to which the first probability distribution is transformed into the second probability distribution; Obtaining a prediction accuracy of the second prediction model; When the accuracy of the second prediction model is equal to or greater than a threshold, the process is terminated. when the accuracy of the second prediction model does not reach or exceed a threshold, a degree to which the first probability distribution is transferred to the second probability distribution is changed, and then the first probability distribution of the first parameter is generated again. Prediction system.

2. The prior distribution generation unit performs cross-validation using the second composition group data to change the degree to which the first probability distribution is transferred to the second probability distribution so as to improve accuracy of the second prediction model. The prediction system of claim 1 .

3. the statistical value of the second probability distribution is a mean value of the second probability distribution; The prediction system of claim 1 .

4. The statistics calculation unit calculates a mean value of the second probability distribution by performing sampling. The prediction system of claim 3 .

5. The statistical calculation unit performs the sampling by a Markov chain Monte Carlo method (MCMC). The prediction system of claim 4.

6. the data acquisition unit acquires third composition group data including a plurality of pairs of composition information and performance values ​​of compositions different from the first composition group data and the second composition group data; the statistics calculation unit calculates, using the second probability distribution as a new prior distribution, statistics of a third probability distribution, which is a posterior probability distribution of observed data of the third composition group data, as parameters of a third prediction model obtained by updating the second prediction model; The prediction system of claim 1 .

7. the first prediction model and the second prediction model are regression models; The parameters are regression coefficients. The prediction system of claim 1 .

8. The number of pairs included in the first composition group data is three times or more the number of pairs included in the second composition group data. The prediction system of claim 1 .

9. The data acquisition unit acquires target composition data, which is composition information of a target composition to be predicted; a prediction unit that inputs composition information of the target composition into the second prediction model to predict a performance value of the target composition; The prediction system of claim 1 further comprising:

10. The performance values ​​are within the range of products that can be produced with the composition. The prediction system of claim 1 .

11. The performance value is the minimum dimension of a product that can be produced from the composition. The prediction system of claim 10.

12. The composition is a resin composition, The composition information includes information on polymers, monomers, and other components in the resin composition. The prediction system of claim 1 .

13. The prediction system according to any one of claims 1 to 12, A model acquisition step of acquiring a first prediction model that predicts a performance value from composition information, the first prediction model being trained based on first composition group data including a plurality of pairs of composition information and performance values ​​of compositions; A data acquisition step of acquiring second composition group data including a plurality of pairs of composition information and performance values ​​of compositions different from the first composition group data; A calculation step of calculating parameters of the second prediction model based on the first prediction model and the second composition group data; A prediction method comprising:

14. The method is executed by a computer, causing the computer to: The prediction system according to any one of claims 1 to 12 is operated as such. program.

Images