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

Abstract

An information processing system according to the present disclosure comprises: an acquisition unit that acquires a content program for a first platform and an application programming interface (API) header pertaining to a second platform different from the first platform; and a generation unit that generates a connection program pertaining to the second platform on the basis of the acquired content program for the first platform and the API header pertaining to the second platform.

Description

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

[0001] The present disclosure relates to an information processing system, an information processing method, and an information processing program.

[0002] Content creators of games, apps, and other content are demanding technology that allows the same content to be used on multiple platforms, including personal computers (PCs), head-mounted displays (HMDs), etc. In the past, making the same content available on multiple platforms required manual work such as porting the content to other platforms or implementing an abstraction layer that makes it easier to use the content on other platforms.

[0003] However, because manual work is costly, there is a demand for technologies that allow the same content to be used on multiple platforms by other means. For example, one such technology is known, which generates content programs for the target platform based on a pre-trained model (see, for example, Patent Document 1).

[0004] Japanese Patent Application Laid-Open No. 2021-174426

[0005] Conventional technology can automatically generate content programs for the target platform, making it possible to make content available on many platforms without manual labor.

[0006] However, generating a content program for a target platform usually requires training a model with a large amount of training data to analyze the characteristics of the platform, which is costly as with manual porting. As such, generating a content program for a target platform may not be able to convert content for one platform into content for another platform at low cost.

[0007] Therefore, an object of the present disclosure is to propose an information processing system, an information processing method, and an information processing program that can convert content for one platform into content for another platform at low cost.

[0008] The information processing system according to the present disclosure includes an acquisition unit that acquires a content program for a first platform and an API (Application Programming Interface) header for a second platform different from the first platform, and a generation unit that generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform.

[0009] 1 is a diagram for explaining an overview of an information processing system according to an embodiment. FIG. 1 is a diagram for explaining an example of a program configuration before and after inter-platform porting work. FIG. 2 is a diagram for explaining an example of a means for converting a connection program related to a porting source platform into a connection program for a porting destination platform. FIG. 2 is a diagram for explaining an example of a configuration of an information processing system according to an embodiment. FIG. 3 is a diagram for explaining an overview of processing by a base model developer terminal, a PF vendor terminal, a content creator terminal, and an end user terminal. FIG. 3 is a block diagram for explaining an example of a configuration of a base model developer terminal. FIG. 4 is a block diagram for explaining an example of a configuration of a PF vendor terminal. FIG. 5 is a block diagram for explaining an example of a configuration of a content creator terminal. FIG. 6 is a block diagram for explaining an example of a configuration of an information processing device according to an embodiment. FIG. 7 is a flowchart for explaining an example of a processing flow by a base model developer terminal. FIG. 8 is a flowchart for explaining an example of a processing flow by a PF vendor terminal. FIG. 9 is a flowchart for explaining an example of a processing flow by a content creator terminal. FIG. 10 is a flowchart for explaining an example of a processing flow by an information processing device. FIG. 11 is a hardware configuration diagram showing an example of a computer that realizes the functions of an information processing device according to an embodiment.

[0010] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the following embodiments, the same components are denoted by the same reference numerals, and redundant description will be omitted.

[0011] Hereinafter, embodiments of the present disclosure will be described in the following order: 1. Embodiment 1-1. Overview of information processing system according to embodiment 1-2. Configuration of information processing system according to embodiment 1-3. Flow of information processing according to embodiment 2. Other embodiments 3. Effects of information processing system according to the present disclosure 4. Hardware configuration 5. Supplementary information

[0012] (1. Embodiment) (1-1. Overview of Information Processing System According to Embodiment) An overview of an information processing system 1 according to an embodiment will be described using Fig. 1. Fig. 1 is a diagram for explaining the overview of an information processing system according to an embodiment.

[0013] The information processing system 1 includes an information processing device 100. The information processing device 100 is a server or the like. The information processing device 100 converts a PF-A content package 10, which is content for a source platform called PF (Platform)-A, into a PF-B content package 20, which is content for a destination platform called PF-B.

[0014] For example, the information processing device 100 converts a PF-A content package 10, which is a Windows (registered trademark) PC game, into a PF-B content package 20, which is an HMD game such as a VR (Virtual Reality) goggle game.

[0015] A platform is a device, operating system, browser, etc. that serves as the foundation for operating an apparatus, etc. Examples of a platform include devices such as a PC or HMD, operating systems such as Windows (registered trademark) or Mac (registered trademark) OS, and browser software provided by Google (registered trademark) or Microsoft (registered trademark).

[0016] Next, an example of the configuration of a program before and after inter-platform porting work, ie, a PF-A content package 10 and a PF-B content package 20, will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of the program configuration before and after inter-platform porting work.

[0017] The PF-A content package 10 comprises a content program 11 and a PF-A connection program 12. Outside the PF-A content package 10, a PF-A API header 13 and a PF-A function program 14 exist.

[0018] The content program 11 is the main program of the PF-A content package 10. For example, the content program 11 is a game program.

[0019] The PF-A connection program 12 is a program that connects the content program 11 with a PF-A API header 13, which is sample code that defines the PF-A API, etc. For example, the PF-A connection program 12 is a connection program for Windows.

[0020] The PF-A API is a set of rules for defining procedures and data for using PF-A functions managed by the PF-A function program 14. For example, the PF-A API is a Windows API for using Windows functions managed by the Windows operating system, which is the PF-A function program 14.

[0021] The PF-A connection program 12 uses the PF-A functions managed by the PF-A function program 14 by calling the PF-A API defined by the PF-A API header 13 from the content program 11 .

[0022] The PF-B content package 20 comprises a content program 11 and a PF-B connection program 21. Outside the PF-B content package 20, a PF-B API header 22 and a PF-B function program 23 exist.

[0023] The PF-B connection program 21 is a program that connects the content program 11 with a PF-B API header 22 that defines the PF-B API. For example, the PF-B connection program 21 is a connection program for an HMD.

[0024] The PF-B connection program 21 uses the PF-B functions managed by the PF-B function program 23 by calling the PF-B API defined by the PF-B API header 22 from the content program 11. For example, the PF-B connection program 21 uses the HMD device driver, which is a PF-B function, by calling the HMD API, which is the PF-B API.

[0025] As described above, of the two content packages, the PF-A content package 10 and the PF-B content package 20, the only differences are the PF-A connection program 12 and the PF-B connection program 21. Therefore, when the connection program of the porting source is converted into the connection program of the porting destination, the content intended for the porting source platform is converted into the content intended for the porting destination platform.

[0026] An example of a means for converting a connection program related to a source platform into a connection program for a destination platform will be described below with reference to Fig. 3. Fig. 3 is a diagram for explaining an example of a means for converting a connection program related to a source platform into a connection program for a destination platform.

[0027] 3, the PF-B connection program 21 is generated based on a machine learning model 200 that outputs the PF-B connection program 21 in response to input of the PF-A connection program 12. In this case, the machine learning model 200 must acquire generalized knowledge regarding the differences between the PF-A connection program 12 and the PF-B connection program 21. Furthermore, acquiring this knowledge requires the preparation of a large number of pairs of learning data for the PF-A connection program 12 and the PF-B connection program 21.

[0028] However, while the training data is available to content creators, it is often not publicly available and is therefore difficult to obtain. Therefore, preparing a large amount of training data requires manual creation, which is costly.

[0029] To solve this problem of high costs, the information processing device 100 according to the embodiment acquires a content program for a first platform and an API header for a second platform different from the first platform. The information processing device 100 also generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform.

[0030] An example of the above-described processing of the information processing device 100 according to the embodiment will be described below with reference to Fig. 4. Fig. 4 is a diagram for explaining an example of processing of the information processing device according to the embodiment. First, the information processing device 100 acquires the content program 11 for PF-A corresponding to the content program for the above-described first platform, and the PF-B API header 22 corresponding to the API header related to the above-described second platform.

[0031] For example, the information processing device 100 can easily obtain the content program 11 by acquiring the content program 11 previously extracted from the PF-A content package 10. Furthermore, the information processing device 100 can obtain sample code that uses the PF-B API as the PF-B API header 22, and although this depends on the prevalence of PF-B, it can relatively easily obtain sample code for a certain PF.

[0032] Next, the information processing device 100 generates a PF-B connection program 21 corresponding to the above-mentioned second connection program based on the acquired content program 11 and PF-B API header 22 .

[0033] For example, the information processing device 100 uses, as the pre-trained machine learning model, a pre-trained language model that receives as input the content program 11 and the PF-B API header 22. The language model is, for example, an LLM (Large Language Model).

[0034] Specifically, the information processing device 100 uses a language model that has been pre-trained to learn how to call the PF-B API header 22 from the content program 11. The information processing device 100 generates a connection program, as the PF-B connection program 21, that connects the content program 11 predicted by the language model with the PF-B API header 22.

[0035] Next, the information processing device 100 connects the content program 11 with the generated PF-B connection program 21, thereby generating a PF-B content package 20, which is content for the second platform to which the content is to be ported.

[0036] In this way, instead of the difficult-to-obtain PF-A connection program 12, the information processing device 100 automatically generates the PF-B connection program 21 based on the easily obtainable content program 11 and the PF-B API header 22. This allows the information processing device 100 to easily generate the PF-B content package 20 based on the PF-B connection program 21 generated at low cost, thereby enabling content for one platform to be converted into content for another platform at low cost.

[0037] Furthermore, when the information processing device 100 uses a pre-trained machine learning model, it is sufficient to pre-train the machine learning model on how to call the PF-B API header 22 from the content program 11. In this case, unlike when the information processing device 100 learns the difference between the PF-A connection program 12 and the PF-B connection program 21, the information processing device 100 does not need to use a large amount of training data for training. This also allows the information processing device 100 to convert content for one platform into content for another platform at low cost.

[0038] (1-2. Configuration of Information Processing System According to Embodiment) Next, an example of the configuration of the information processing system 1 according to the embodiment will be described with reference to FIG. 5. FIG. 5 is a diagram showing an example of the configuration of the information processing system according to the embodiment. In FIG. 5, the information processing system 1 includes an information processing device 100 and an end-user terminal 300.

[0039] 5, the information processing device 100 trains a machine learning model such as a language model, and then generates the PF-B connection program 21 based on the pre-trained machine learning model. The information processing device 100 also includes, for example, a base model developer terminal 100A, a PF vendor terminal 100B, and a content creator terminal 100C.

[0040] The base model developer terminal 100A, the PF vendor terminal 100B, and the content creator terminal 100C all train machine learning models. The end user terminal 300 does not train machine learning models.

[0041] The base model developer terminal 100A is a terminal on the developer side of the API usage program automatic generation model, which is a machine learning model that forms the basis of the machine learning model that generates the PF-B connection program 21.

[0042] For example, the base model developer terminal 100A generates a language model that has been pre-trained on how to call a general-purpose API header from a general-purpose content program as an API-using program automatic generation model. The base model developer terminal 100A provides the API-using program automatic generation model to the PF vendor terminal 100B.

[0043] The PF vendor terminal 100B is a terminal on the developer side of PF-B. For example, the PF vendor terminal 100B is a terminal on the developer side of an HMD. The PF vendor terminal 100B generates a PF connection program generation model, which is a machine learning model obtained by modifying and additionally learning the API usage program automatic generation model.

[0044] As one example, the PF vendor terminal 100B generates, as a PF connection program generation model, how to call the PF-B API header 22 from the content program 11. As another example, the PF vendor terminal 100B generates a language model that has been pre-trained with knowledge related to a connection program that connects the content program 11 and the PF-B API header 22. The PF vendor terminal 100B provides the PF connection program generation model to the content creator terminal 100C.

[0045] The content creator terminal 100C performs additional learning on the PF connection program generation model until the content to be ported to PF-B operates normally on PF-B. The content creator terminal 100C ports the content to PF-B based on the PF connection program generation model for which additional learning has been completed. The content creator terminal 100C provides the content ported to PF-B to the end user terminal 300.

[0046] (Processing Overview) Next, an overview of the processing by the base model developer terminal 100A, the PF vendor terminal 100B, the content creator terminal 100C, and the end user terminal 300 will be described with reference to Fig. 6. Fig. 6 is a diagram for explaining an overview of the processing by the base model developer terminal, the PF vendor terminal, the content creator terminal, and the end user terminal.

[0047] The content creator terminal 100C acquires PC content, such as a PC game, corresponding to the PF-A content package 10 created by the content creator (step S1). The content creator terminal 100C uploads the PC content to a content porting service network operated by the PF vendor terminal 100B, which is a terminal for HMD vendors (step S2).

[0048] The PF vendor terminal 100B transfers the PC content from the PC, which is PF-A, to the HMD corresponding to PF-B (step S3). For example, the PF vendor terminal 100B generates HMD content such as an HMD game based on a PF connection program generation model that has been modified and additionally learned from the API usage program automatic generation model provided from the base model developer terminal 100A.

[0049] Specifically, the PF vendor terminal 100B generates an HMD connection program by inputting the game program and the API header of the HMD into the PF connection program generation model that has undergone additional learning, etc. Then, the PF vendor terminal 100B generates an HDM game by connecting the HMD connection program and the game program.

[0050] The PF vendor terminal 100B uploads the generated HMD content to the content porting service network. In this case, the PF vendor terminal 100B can also upload a PF connection program generation model to the network. The content creator terminal 100C downloads the HMD content from the network (step S4). In this case, the content creator terminal 100C can also download the PF connection program generation model.

[0051] The content creator terminal 100C provides the end user terminal 300 with the HMD content downloaded in step S4 in addition to the PC content acquired in step S1.

[0052] For example, the content creator terminal 100C releases a PC game to the PC gamer terminal 300A, and the content creator terminal 100C releases an HMD game to the HMD gamer terminal 300B.

[0053] When the content creator terminal 100C downloads a PF connection program generation model, it can also release HMD content generated based on the additionally trained PF connection program generation model to the HMD gamer terminal 300B.

[0054] Steps S1 to S5 enable content to be ported at low cost to HMDs, which have less content than PCs. Games that use 3D (Dimension) and CG (Computer Graphics) among such content can be ported at low cost regardless of the platform, such as a PC or HMD, because only the connection program described above is different.

[0055] (Configuration of the base model developer terminal) Next, an example of the configuration of the base model developer terminal 100A will be described using Fig. 7. Fig. 7 is a block diagram showing an example of the configuration of the base model developer terminal. In Fig. 7, the base model developer terminal 100A includes an API-using program automatic generation processing unit 101A, a compiler 102A, and a machine language instruction consistency verification unit 103A.

[0056] The API-using program automatic generation processor 101A is an API-using program automatic generation model. The API-using program automatic generation processor 101A is, for example, a language model based on Transformer. The API-using program automatic generation processor 101A includes a machine language instruction token sequencing unit 1011, a connection program self-attention unit 1012, a program feature vectorization unit 1013, an API cross-attention unit 1014, and a machine language instruction token sequence prediction unit 1015.

[0057] The machine language instruction token sequencing unit 1011 corresponds to a Transformer decoder. The machine language instruction token sequencing unit 1011 receives an input of a content program+connection program 21A1. The content program+connection program 21A1 is training data that includes a content program that uses a general-purpose API header 22A and a general-purpose connection program for which a PF-B is not specified.

[0058] For example, the machine language instruction token series generation unit 1011 receives a large amount of input of a content program+connection program 21A1 that includes a content program, which is an application or source code that uses the definition of an API present on the Internet, and a connection program.

[0059] In FIG. 7, the machine-language instruction token sequencing unit 1011 receives input of content machine-language instructions 21A1X obtained by converting the content program+connection program 21A1 into machine-language instructions by the compiler 102A.

[0060] Next, the machine language instruction token series generation unit 1011 converts the content machine language instruction 21A1X, which is a series of consecutive characters or symbols, into a feature vector (or embedding vector) series for each token, as in natural language.

[0061] The connection program Self Attention unit 1012 learns the relationship between the elements of the machine language instruction token sequence, which is a feature vector sequence converted from the content machine language instruction 21A1X.

[0062] The program feature vectorization unit 1013 corresponds to the Encoder of a Transformer. The program feature vectorization unit 1013 receives input of a general-purpose API header 22A for which a PF-B is not specified as training data. For example, the program feature vectorization unit 1013 receives a large amount of input of definitions of APIs available on the Internet. Next, the program feature vectorization unit 1013 converts the API header 22A, which is a sequence of consecutive characters or symbols, into a feature vector sequence for each token.

[0063] The API cross attention unit 1014 learns the relationship between each element of the first program feature vector, which is a feature vector sequence converted from the API header 22A, and each element of the machine language instruction token sequence. In this way, the API cross attention unit 1014 learns how to call the API header 22A from the content program of the content program+connection program 21A1.

[0064] The machine language instruction token sequence prediction unit 1015 predicts the next token in the machine language instruction token sequence based on how the API header 22A is called from the content program of the content program+connection program 21A1. The machine language instruction token sequence prediction unit 1015 also converts the machine language instruction token sequence and the predicted next token from a feature vector sequence into a machine language instruction, thereby generating a content machine language instruction 21A2X.

[0065] The machine language instruction match verification unit 103A verifies whether the content machine language instructions 21A1X input to the API-using program automatic generation processing unit 101A match the content machine language instructions 21A2X output from the API-using program automatic generation processing unit 101A.

[0066] If the content machine instructions 21A1X and 21A2X do not match, the machine instruction match verification unit 103A feeds back the discrepancy to the API-using program automatic generation processing unit 101A until they substantially match. This causes the machine instruction match verification unit 103A to repeatedly update the parameters of the API-using program automatic generation processing unit 101A. In this way, the machine instruction match verification unit 103A causes the API-using program automatic generation processing unit 101A to perform additional learning (supervised learning) using the error backpropagation method.

[0067] In this case, the machine language instruction matching verification unit 103A may, for example, cause each unit of the API-using program automatic generation processing unit 101A to perform additional learning, or may cause only the Cross Attention unit for API, such as the Cross Attention unit for API 1014, to perform additional learning.

[0068] (Configuration of PF Vendor Terminal) Next, an example of the configuration of the PF vendor terminal 100B will be described using Fig. 8. Fig. 8 is a block diagram showing an example of the configuration of the PF vendor terminal. Compared to the base model developer terminal 100A, the PF vendor terminal 100B includes an automatic connection program generation processing unit 101B and a machine instruction match verification unit 103B instead of the API-using program automatic generation processing unit 101A and the machine instruction match verification unit 103A.

[0069] The automatic connection program generation processing unit 101B is a PF connection program generation model in which modifications and additional learning have been performed on the API-using program automatic generation model. The automatic connection program generation processing unit 101B further includes a program feature vectorization unit 1016 and a content cross attention unit 1017 in addition to the automatic API-using program generation processing unit 101A.

[0070] In addition, the automatic connection program generation processing unit 101B includes a machine language instruction token sequencing unit 1011B, a program feature vectorization unit 1013B, and a machine language instruction token sequence prediction unit 1015B, instead of the machine language instruction token sequencing unit 1011, the program feature vectorization unit 1013, and the machine language instruction token sequence prediction unit 1015.

[0071] The machine-language instruction token sequencing unit 1011B receives input of a PF-B connection program 21B1 that connects the content program 11A with the PF-B API header 22A, instead of the content program+connection program 21A1. In Fig. 8, the machine-language instruction token sequencing unit 1011B receives input of connection machine-language instructions 21B1X that have been converted from the PF-B connection program 21B1 by the compiler 102A.

[0072] The program feature vectorization unit 1013B accepts input of a PF-B API header 22B as training data instead of a general-purpose API header 22A that does not specify a PF-B. For example, the program feature vectorization unit 1013B accepts input of an API definition of a PF-B (such as an HMD) that is a PF managed by the PF vendor itself.

[0073] Next, the program feature vectorization unit 1013B converts the PF-B API header 22B, which is a series of consecutive characters or symbols, into a feature vector series for each token, thereby obtaining a first program feature vector.

[0074] The program feature vectorization unit 1016 corresponds to the Encoder of a Transformer. The program feature vectorization unit 1016 receives input of a content program 11A that uses the PF-B API header 22A as training data. For example, the program feature vectorization unit 1016 receives input of a content program 11A created by a PF vendor or a content program 11A collected from the Internet.

[0075] Next, the program feature vectorization unit 1016 converts the content program 11A, which is a series of consecutive characters or symbols, into a feature vector series for each token.

[0076] The content cross attention unit 1017 learns the relationship between each element of the second program feature vector, which is a feature vector sequence converted from the content program 11A, each element of the first program feature vector, and each element of the machine language instruction token sequence. In this way, the content cross attention unit 1017 learns not just how to call the general-purpose API header 22A, but also how to call the PF-B API header 22B and knowledge related to the connection program that connects the content program 11A and the API header 22B.

[0077] The machine language instruction token sequence predictor 1015B converts the machine language instruction token sequence and the predicted next token from the feature vector sequence into a machine language instruction, thereby generating a concatenated machine language instruction 21B2X.

[0078] The machine instruction match verification unit 103B verifies whether the connecting machine instruction 21B1X input to the automatic connecting program generation processing unit 101B matches the connecting machine instruction 21B2X output from the automatic connecting program generation processing unit 101B.

[0079] If the connecting machine instruction 21B1X and the connecting machine instruction 21B2X do not match, the machine instruction match verification unit 103B feeds back these errors to the automatic connecting program generation processor 101B until they approximately match. This causes the machine instruction match verification unit 103B to repeatedly update the parameters of the automatic connecting program generation processor 101B. In this way, the machine instruction match verification unit 103B causes the automatic connecting program generation processor 101B to perform additional learning (supervised learning) using the error backpropagation method.

[0080] (Configuration of the Content Creator Terminal) Next, an example of the configuration of the content creator terminal 100C will be described with reference to Fig. 9. Fig. 9 is a block diagram showing an example of the configuration of the content creator terminal.

[0081] The content creator terminal 100C includes an automatic connection program generation processor 101C instead of the automatic connection program generation processor 101B of the PF vendor terminal 100B.

[0082] Furthermore, the content creator terminal 100C includes a compiler 102C, a linker 104C, a link error verification unit 105C, and a unit test verification unit 106C, instead of the compiler 102A and the machine language instruction agreement verification unit 103B. This is because the content creator terminal 100C performs reinforcement learning without using teacher data, instead of supervised learning using teacher data that is the target for comparing the success or failure of data output from the language model. Reinforcement learning by the link error verification unit 105C and the unit test verification unit 106C will be described later.

[0083] The connection program automatic generation processing unit 101C is a language model that has undergone additional learning of the PF connection program generation model until the content to be ported to PF-B operates normally on PF-B.

[0084] The automatic connection program generation processing unit 101C includes a machine language instruction token sequencing unit 1011C, a program feature vectorization unit 1013C, a program feature vectorization unit 1013C, and a machine language instruction token sequence prediction unit 1015C, instead of the machine language instruction token sequencing unit 1011B, the program feature vectorization unit 1013B, the program feature vectorization unit 1016B, and the machine language instruction token sequence prediction unit 1015B.

[0085] The machine language instruction token series generating unit 1011C receives input of a connecting machine language instruction start token 21C1X for which no machine language instruction has been generated, instead of a connecting machine language instruction 21B1X for which a machine language instruction has been generated in advance.

[0086] The program feature vectorization unit 1013C receives an input of the PF-B API header 22C. The PF-B API header 22C is the same as the PF-B API header 22B in that it contains only the API header of the PF-B platform to which the program is ported. However, the PF-B API header 22B differs from the PF-B API header 22B in that it is not training data for use in supervised learning.

[0087] The program feature vectorization unit 1016C receives input of the content program 11C. The content program 11C differs from the content program 11B in that it is only a program of the content that the content creator wants to port and that it is not training data to be used in supervised learning.

[0088] The machine language instruction token sequence prediction unit 1015C converts the machine language instruction token sequence and the predicted next token from a feature vector sequence into a machine language instruction, thereby generating a connection machine language instruction 21C2X, which is a machine language instruction corresponding to the PF-B connection program 21.

[0089] The compiler 102C converts the content program 11C and the PF-B API header 22C into machine language instructions. The linker 104C links (connects) the generated connecting machine language instructions 21C2X with the content program 11C and the PF-B API header 22C that have been converted into machine language instructions.

[0090] The link error verification unit 105C verifies whether a link error occurs in the link between the generated connecting machine language instruction 21C2X, the content program 11C converted into the machine language instruction, and the PF-B API header 22C. For example, the link error verification unit 105C verifies that a link error has occurred if the PF-B PI header called by the connecting machine language instruction 21C2X is different from the PF-B API header 22C.

[0091] The link error verification unit 105C feeds back the loss function to the automatic connection program generation processing unit 101C until no link errors occur (until the loss function based on link errors is minimized). As a result, the link error verification unit 105C repeatedly updates the parameters of the automatic connection program generation processing unit 101C. In this way, the link error verification unit 105C causes the automatic connection program generation processing unit 101C to perform additional learning (reinforcement learning) using the backpropagation method.

[0092] The unit test verification unit 106C verifies whether all functions of the content that has at least become executable by passing the link error verification unit 105C operate normally using an automatic test prepared in advance by the content creator. For example, the unit test verification unit 106C verifies whether all functions of the content operate normally by comparing the executable content with a content unit test 24C, which is content for automatic testing prepared in advance by the content creator.

[0093] The unit test verification unit 106C feeds back the loss function to the automatic connection program generation processing unit 101C until all the automatic tests pass (until the loss function based on the automatic tests is minimized). As a result, the unit test verification unit 106C repeatedly updates the parameters of the automatic connection program generation processing unit 101C. In this way, the unit test verification unit 106C causes the automatic connection program generation processing unit 101C to perform additional learning (reinforcement learning) using the backpropagation method.

[0094] According to the above learning, the finally generated PF-B content package 20 can be made to operate on the PF-B as expected by the content creator.

[0095] (Configuration of Information Processing Device) In the above example, the base model developer terminal 100A, the PF vendor terminal 100B, and the content creator terminal 100C are configured as separate devices. However, these can also be configured conceptually or schematically as a single information processing device 100. An example of the configuration of the information processing device 100 according to the embodiment will be described below with reference to FIG. 10. FIG. 10 is a block diagram showing an example of the configuration of the information processing device according to the embodiment.

[0096] The information processing device 100 includes a communication unit 110, a storage unit 120, and a control unit 130. The communication unit 110 is realized by, for example, a network interface controller or the like. The communication unit 110 is connected to a network N such as the Internet via a wired or wireless connection, and transmits and receives information via the network N.

[0097] For example, the communication unit 110 transmits and receives information using communication standards and technologies such as Wi-Fi (registered trademark), SIM (Subscriber Identity Module), LPWA (Low Power Wide Area), etc. Specifically, the communication unit 110 receives a content program for a first platform and an API header related to a second platform.

[0098] The storage unit 120 is realized by, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 120 stores various data. For example, the storage unit 120 stores a language model.

[0099] The control unit 130 is realized, for example, by a central processing unit (CPU), a micro processing unit (MPU), or the like executing a program (for example, an information processing program according to the present disclosure) stored inside the information processing device 100 using a random access memory (RAM) or the like as a work area. The control unit 130 is a controller, and may be realized, for example, by an integrated circuit such as an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

[0100] The control unit 130 includes a learning unit 131, an acquisition unit 132, and a generation unit 133. The learning unit 131 pre-trains a machine learning model such as a language model. For example, the learning unit 131 trains an API-using program automatic generation processing unit 101A, an automatic connection program generation processing unit 101B, and an automatic connection program generation processing unit 101C.

[0101] The acquisition unit 132 acquires a content program for a first platform and an API header for a second platform different from the first platform. For example, the acquisition unit 132 acquires the content program 11 and the PF-B API header 22.

[0102] Furthermore, the acquisition unit 132 acquires, as the API header for the second platform, an API header for a second platform for which the number of available contents is smaller than that of the first platform. For example, the acquisition unit 132 acquires, as the PF-B API header 22, an API header for an HMD for which the number of available contents is smaller than that of a PC.

[0103] The generation unit 133 generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform. For example, the generation unit 133 generates the PF-B connection program 21 by inputting the acquired content program 11 and the PF-B API header 22 into a pre-trained language model.

[0104] Furthermore, the generation unit 133 uses, as the pre-trained language model, a pre-trained language model that receives a general-purpose content program and a general-purpose API header as input. For example, the generation unit 133 uses, as the pre-trained language model, the API-using program automatic generation processing unit 101A that has been pre-trained to learn how to call the general-purpose API header 22A from the general-purpose content program included in the content program+connection program 21A1.

[0105] When the language model inputs a general-purpose content program, the generation unit 133 uses an API-utilizing program automatic generation processing unit 101A that inputs content machine language instructions 21A1X, which are content programs converted into machine language instructions, as the language model.

[0106] Furthermore, the generation unit 133 uses, as the pre-trained language model, a language model that has been subjected to supervised learning so that a general-purpose connection program input to the language model and a general-purpose connection program output from the language model substantially coincide with each other. For example, the generation unit 133 uses the API-using program automatic generation processing unit 101A that has been subjected to supervised learning so that a content machine language instruction 21A1X and a content machine language instruction 21A2X substantially coincide with each other.

[0107] Furthermore, the generation unit 133 uses a pre-trained language model that outputs a connection program related to the second platform as the pre-trained language model. For example, when the generation unit 133 outputs a connection program related to the second platform as the pre-trained language model, the generation unit 133 uses the connection program automatic generation processing unit 101B that outputs connection machine language instructions 21B2X, which are connection programs converted into machine language instructions.

[0108] Furthermore, the generation unit 133 uses, as the pre-trained language model, a language model that has been subjected to supervised learning so that a connection program for the second platform input to the language model and a connection program for the second platform output from the language model substantially coincide with each other. For example, the generation unit 133 uses, as the pre-trained language model, the connection program automatic generation processing unit 101B that has been subjected to supervised learning so that a connection machine language instruction 21B1X and a connection machine language instruction 21B2X substantially coincide with each other.

[0109] Furthermore, the generation unit 133 uses a pre-trained language model that receives an API header related to the second platform as input as the pre-trained language model. For example, the generation unit 133 uses the connection program automatic generation processing unit 101B that receives a PF-B API header 22B as input as the pre-trained language model.

[0110] In addition, the generation unit 133 uses, as a pre-trained language model, a language model that has been pre-trained with at least one of knowledge regarding how to call an API header related to the second platform from a general-purpose content program and knowledge regarding a connection program that connects the general-purpose content program and the API header related to the second platform.

[0111] For example, the generation unit 133 uses the automatic connection program generation processing unit 101B that has previously learned how to call the PF-B API header 22B and knowledge about the connection program that connects the content program 11A and the API header 22B.

[0112] Furthermore, the generation unit 133 uses a pre-trained language model that takes a content program for the first platform as input as the pre-trained language model. For example, the generation unit 133 uses an automatic connection program generation processing unit 101C that takes a content program 11C as input as the pre-trained language model.

[0113] Furthermore, the generation unit 133 uses, as the pre-trained language model, a pre-trained language model that outputs a connection program related to the second platform. For example, the generation unit 133 uses, as the pre-trained language model, the connection program automatic generation processing unit 101C that outputs a connection machine language instruction 21C2X corresponding to the PF-B connection program 21.

[0114] Furthermore, the generation unit 133 uses, as the pre-trained language model, a language model that has been subjected to reinforcement learning based on the verification results of content for the second platform, including a connection program for the second platform, output from the language model. For example, the generation unit 133 uses the connection program automatic generation processing unit 101C that has been subjected to reinforcement learning based on the verification results by the link error verification unit 105C and the unit test verification unit 106C.

[0115] Furthermore, the generation unit 133 uses, as the pre-trained language model, a language model whose parameters have been updated by reinforcement learning until it is verified that no link errors occur in the content for the second platform.The generation unit 133 uses the language model whose parameters have been updated by reinforcement learning until it is verified that no abnormalities occur in the function of the content for the second platform.

[0116] For example, the generation unit 133 uses the connection program automatic generation processing unit 101C in which parameters are updated by reinforcement learning until the loss function based on link errors and the loss function based on automatic testing are minimized.

[0117] Furthermore, the generation unit 133 generates content for the second platform including a connection program for the second platform. For example, the generation unit 133 generates PF-B content in which the content program 11, the PF-B connection program 21, and the PF-B API header 22 are connected.

[0118] Furthermore, the generation unit 133 generates content for the second platform by converting content for a PC corresponding to the first platform into content for an HMD corresponding to the second platform. For example, the generation unit 133 generates a game by converting a PC game corresponding to the PF-A content package 10 into an HMD game corresponding to the PF-B content package 20.

[0119] (1-3. Information processing flow according to embodiment) An example of the information processing flow according to the embodiment will be described using Figs. 11 to 14. Fig. 11 is a flowchart showing an example of the processing flow by the base model developer terminal. Fig. 12 is a flowchart showing an example of the processing flow by the PF vendor terminal. Fig. 13 is a flowchart showing an example of the processing flow by the content creator terminal. Fig. 14 is a flowchart showing an example of the processing flow by the information processing device.

[0120] (Processing Flow by the Base Model Developer Terminal) First, an example of the processing flow by the base model developer terminal 100A will be described using Figure 11. The API-using program automatic generation processing unit 101A accepts input of an API header 22A of any library collected in large quantities in advance, and a content program + connection program 21A1 such as a sample application (step S11).

[0121] The API-using program automatic generation processing unit 101A performs pre-learning of the API-using program automatic generation model (step S12).

[0122] The API-using program automatic generation processing unit 101A learns in response to the input of the API header 22A until it becomes possible to generate a program that uses the API header 22A (step S13). For example, the machine-language instruction match verification unit 103A causes the API-using program automatic generation processing unit 101A to perform supervised learning until the content machine-language instructions 21A1X and the content machine-language instructions 21A2X substantially match.

[0123] The base model developer terminal 100A ships the trained API-using program automatic generation processing unit 101A to the PF vendor terminal 100B as an API-using program automatic generation model (base model) (step S14).

[0124] (Processing Flow by PF Vendor Terminal) Next, an example of the processing flow by the PF vendor terminal 100B will be described with reference to Fig. 12. In the following, the PF vendor terminal 100B performs steps S23 and S24 in parallel with steps S21 and S22.

[0125] The PF vendor terminal 100B acquires an API-using program automatic generation model (base model) (step S21). The connection program automatic generation processing unit 101B activates the content Cross Attention unit 1017 (step S22).

[0126] The connection program automatic generation processing unit 101B receives input of a small amount of existing applications and the like collected in advance as the content program 11A, the PF-B connection program 21B1, and the PF-B API header 22B (step S23).

[0127] The connection program automatic generation processing unit 101B learns knowledge about a program that connects the PF-B API header 22B, which is an API header for the PF vendor's own PF, and the content program 11A, which is an existing application or the like (step S24).

[0128] The automatic connection program generation processing unit 101B performs additional learning of the PF connection program generation model (step S25). For example, if the connection machine instruction 21B1X and the connection machine instruction 21B2X do not match, the machine instruction match verification unit 103B causes the automatic connection program generation processing unit 101B to perform supervised learning.

[0129] The automatic connection program generation processing unit 101B learns, in response to the input of the PF-B API header 22B and the content program 11A, which is an existing application or the like, until it is able to generate a connection program that connects these (step S26). For example, the machine-language instruction match verification unit 103B causes the automatic connection program generation processing unit 101B to perform supervised learning until the connection machine-language instruction 21B1X and the connection machine-language instruction 21B2X match.

[0130] The PF vendor terminal 100B ships the learned connection program automatic generation processing unit 101B to the content creator terminal 100C as a PF connection program generation model (step S27).

[0131] (Processing Flow by Content Creator Terminal) Next, an example of the processing flow by the content creator terminal 100C will be described with reference to Fig. 13. In the following, the content creator terminal 100C performs steps S32 and S33 in parallel with step S31.

[0132] The content creator terminal 100C acquires the PF connection program generation model from the PF vendor terminal 100B (step S31).

[0133] The automatic connection program generation processor 101C generates content for the destination platform (step S32). For example, the automatic connection program generation processor 101C generates an HMD game corresponding to the PF-B content package 20. The unit test verification unit 106C acquires the content unit test 24C (step S33).

[0134] The link error verification unit 105C verifies whether or not a link error occurs (step S34). For example, the link error verification unit 105C verifies whether or not a link error occurs in the link between the connection machine language instruction 21C2X and the content program 11C and PF-B API header 22C converted into the machine language instruction. The link error verification unit 105C causes the connection program automatic generation processing unit 101C to perform reinforcement learning until no link errors occur.

[0135] The unit test verification unit 106C verifies whether all functions of the content operate normally by comparing the content that has passed the link error verification unit 105C with the content unit test 24C (step S35). The unit test verification unit 106C causes the automatic connection program generation processing unit 101C to perform reinforcement learning until the content passes all automatic tests.

[0136] The content creator terminal 100C ships the PF-B content package 20, which is content for the porting platform generated by the trained connection program automatic generation processing unit 101C, to the end user terminal 300 (step S36). For example, the content creator terminal 100C ships an HMD game to the end user terminal 300.

[0137] (Processing Flow by Information Processing Device) Next, an example of the processing flow by the information processing device 100 will be described with reference to Fig. 14. The learning unit 131 pre-trains a machine learning model such as a language model (step S101). For example, the learning unit 131 trains the API-using program automatic generation processing unit 101A, the connection program automatic generation processing unit 101B, and the connection program automatic generation processing unit 101C.

[0138] The acquisition unit 132 acquires a content program for the first platform and an API header for a second platform different from the first platform (step S102). For example, the acquisition unit 132 acquires the content program 11 and the PF-B API header 22.

[0139] The generation unit 133 generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform (step S103). For example, the generation unit 133 generates the PF-B connection program 21 by inputting the acquired content program 11 and the PF-B API header 22 into a pre-trained language model.

[0140] (2. Other Embodiments) The processes according to the embodiments can be implemented in various different forms other than the above-described embodiments.

[0141] Of the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using known methods. In addition, the information including the processing procedures, specific names, various data, and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified. The various information shown in each drawing is not limited to the information shown in the drawings.

[0142] The components of each device shown in the figure are conceptual functional components and do not necessarily have to be physically configured as shown. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of the configuration can be functionally or physically distributed or integrated in any unit depending on various loads and usage conditions.

[0143] The above-described embodiments and modifications can be combined as appropriate within the scope of not causing any contradiction in the processing content.

[0144] The effects described in this specification are merely examples and are not limiting, and other effects may also be present.

[0145] (3. Effects of the information processing system according to the present disclosure) As described above, the information processing system according to the present disclosure (information processing system 1 in the embodiment) includes an acquisition unit (acquisition unit 132 in the embodiment) and a generation unit (generation unit 133 in the embodiment).

[0146] The acquisition unit acquires a content program for a first platform and an API header for a second platform different from the first platform, and the generation unit generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform.

[0147] In this way, instead of a connection program for the first platform that is difficult to obtain, the information processing system automatically generates a connection program for the second platform based on a content program for the first platform that is easy to obtain and an API header for the second platform.

[0148] This allows the information processing system to easily generate content for the second platform based on a connection program for the second platform generated at low cost, thereby enabling content for one platform to be converted into content for another platform at low cost.

[0149] The generation unit generates content for the second platform, including a connection program for the second platform. The information processing system can generate the content for the second platform simply by connecting the content program with the generated connection program for the second platform. This allows the information processing system to convert content for other platforms at low cost.

[0150] The generation unit generates, as the content for the second platform, content that is converted from content for a PC corresponding to the first platform into content for an HMD corresponding to the second platform. In this way, the information processing device can convert content for a PC into content for an HMD at low cost.

[0151] The acquisition unit acquires, as the API header for the second platform, an API header for the second platform that has a smaller number of available contents than the first platform.

[0152] The information processing system can easily generate content for the second platform by obtaining an API header for the second platform, which is readily available, and thus can convert content for one platform into content for another platform at low cost, even if the second platform has fewer available contents than the first platform.

[0153] The generation unit generates a connection program for the second platform based on the pre-trained language model. For example, the information processing system can generate the connection program for the second platform simply by inputting the acquired content program for the first platform and an API header for the second platform into the pre-trained language model. This allows the information processing system to convert content for one platform into content for another platform at low cost.

[0154] The generation unit uses, as the pre-trained language model, a pre-trained language model that receives a general-purpose content program and a general-purpose API header as input.

[0155] By using a pre-trained language model based on a general-purpose content program and a general-purpose API header as input, the information processing system can easily perform appropriate modifications and additional training on the base language model. Therefore, by using a pre-trained language model based on the language model, the information processing system can convert content for one platform into content for another platform at lower cost.

[0156] The generation unit uses, as the pre-trained language model, a language model that has been pre-trained on how to call a general-purpose API header from a general-purpose content program.

[0157] By using a language model that has been pre-trained on how to call a general-purpose API header from a general-purpose content program as a base, the information processing system can more easily appropriately modify and further train the base language model. Therefore, by using a pre-trained language model based on the language model, the information processing system can convert content for one platform into content for another platform at lower cost.

[0158] When a general-purpose content program is input as the pre-trained language model, the generation unit uses a pre-trained language model that receives as input a content program converted into machine language instructions.

[0159] A content program converted into machine language instructions has less freedom in describing the same process than a content program in source code form that has not been converted into machine language instructions. Therefore, by using a pre-trained language model that inputs a content program converted into machine language instructions, the information processing system can reduce variations in the connection program generated for the second platform. This allows the information processing system to improve learning efficiency.

[0160] The generation unit uses, as the pre-trained language model, a language model that has been subjected to supervised learning so that a general-purpose connection program input to the language model and a general-purpose connection program output from the language model approximately match.

[0161] The information processing system uses a language model that has been subjected to supervised learning until these general-purpose connection programs are approximately identical, thereby generating a program that uses the API header of the second platform in response to the input of the API header. Therefore, by using the language model, the information processing system can generate a connection program for the second platform that is appropriate as a program that uses the API header.

[0162] The generation unit uses a pre-trained language model that outputs a connection program for the second platform as the pre-trained language model. This allows the information processing system to generate the connection program for the second platform from the output of the language model, thereby enabling content for one platform to be converted into content for another platform at lower cost.

[0163] When the generation unit outputs a connection program for the second platform as a pre-trained language model, the generation unit uses a pre-trained language model that outputs a connection program for the second platform converted into machine language instructions.

[0164] A connection program for the second platform that has been converted into machine language instructions has less freedom in describing the same processing than a connection program for the second platform in the form of source code that has not been converted into machine language instructions.

[0165] Therefore, by using a pre-trained language model that takes as input a connection program for the second platform converted into machine language instructions, the information processing system can reduce variations in the connection program for the second platform that is generated, thereby improving the learning efficiency of the information processing system.

[0166] The generation unit uses, as the pre-trained language model, a language model that has been subjected to supervised learning so that the connection program for the second platform input to the language model and the connection program for the second platform output from the language model are approximately identical.

[0167] The information processing system can generate a connection program for connecting the API header and content program of the second platform in response to input of the API header and content program of the second platform by using a language model that has been subjected to supervised learning until the connection program approximately matches the API header and content program of the second platform. Therefore, by using the language model, the information processing system can generate a connection program for the second platform that is appropriate as a connection program for connecting the API header and content program of the second platform.

[0168] The generation unit uses a pre-trained language model that receives an API header related to the second platform as input. This allows the information processing system to use a pre-trained language model that is the same as the acquired API header related to the second platform. Therefore, in response to the input of the API header and content program of the second platform, the information processing system can generate a connection program for the second platform that is appropriate as a connection program for connecting them.

[0169] The generation unit uses, as a pre-trained language model, a language model that has been pre-trained with at least one of knowledge regarding how to call an API header related to the second platform from a general-purpose content program and knowledge regarding a program that connects the general-purpose content program with an API header related to the second platform.

[0170] As a result, the information processing system generates a connection program for the second platform based on the calling method and the knowledge that has been pre-trained in the language model. As a result, the information processing system can generate a connection program for the second platform that is appropriate as a connection program for connecting the API header and content program of the second platform, in accordance with the input of the API header and content program of the second platform.

[0171] The generation unit uses a pre-trained language model that takes a content program for the first platform as input as the pre-trained language model. This allows the information processing system to use the same pre-trained language model as the acquired content program for the first platform. Therefore, in response to input of the API header and content program of the second platform, the information processing system can generate a connection program for the second platform that is appropriate as a connection program for connecting them.

[0172] The generation unit uses, as the pre-trained language model, a language model that has been reinforced learned based on the verification results of content for the second platform, including a connection program for the second platform, output from the language model. This allows the information processing system to use the language model that has been reinforced learned based on the verification results of content for the second platform, thereby enabling the generation of appropriately verified content for the second platform.

[0173] The generation unit uses, as the pre-trained language model, a language model whose parameters have been updated by reinforcement learning until it is verified that no link errors will occur in the content for the second platform. This allows the information processing system to use the language model that has been reinforced learned until it is verified that no link errors will occur, thereby enabling the generation of content for the second platform that is free of link errors.

[0174] The generation unit uses, as the pre-trained language model, a language model whose parameters have been updated by reinforcement learning until it is verified that no abnormalities occur in the functionality of the content for the second platform. This allows the information processing system to use the language model that has been reinforced learned until it is verified that no abnormalities occur in the functionality of the content for the second platform, without requiring training data. This allows the information processing system to generate content for the second platform that has no abnormalities in function at low cost.

[0175] (4. Hardware Configuration) An information device such as the information processing device 100 according to the embodiment is realized by a computer 1000 configured as shown in Fig. 15. Fig. 15 is a hardware configuration diagram showing an example of a computer that realizes the functions of the information processing device according to the embodiment. The computer 1000 includes a CPU 1100, a RAM 1200, a ROM 1300, a HDD (Hard Disk Drive) 1400, a communication interface 1500, and an input / output interface 1600. The components of the computer 1000 are connected by a bus 1050.

[0176] The CPU 1100 operates and controls each unit based on programs stored in the ROM 1300 or the HDD 1400. The CPU 1100 loads the programs stored in the ROM 1300 or the HDD 1400 into the RAM 1200 and executes processing corresponding to the various programs.

[0177] The ROM 1300 stores boot programs such as a Basic Input Output System (BIOS) that is executed by the CPU 1100 when the computer 1000 starts up, and programs that depend on the hardware of the computer 1000 .

[0178] HDD 1400 is a computer-readable recording medium that non-temporarily records programs executed by CPU 1100 and data used by such programs. Specifically, HDD 1400 is a recording medium that records an information processing program according to the present disclosure, which is an example of program data 1450.

[0179] The communication interface 1500 is an interface for connecting the computer 1000 to an external network 1550 (such as the Internet). The CPU 1100 receives data from other devices and transmits data generated by the CPU 1100 to other devices via the communication interface 1500.

[0180] The input / output interface 1600 is an interface for connecting the input / output device 1650 and the computer 1000. The CPU 1100 receives data from input devices such as a keyboard and a mouse via the input / output interface 1600. The CPU 1100 transmits data to output devices such as a display, a speaker, and a printer via the input / output interface 1600. The input / output interface 1600 can also function as a media interface for reading a program recorded on a predetermined recording medium.

[0181] The media may be optical recording media such as DVDs (Digital Versatile Discs) and PDs (Phase Change Rewritable Discs), magneto-optical recording media such as MOs (Magneto-Optical disks), tape media, magnetic recording media, or semiconductor memories.

[0182] When the computer 1000 functions as the information processing device 100 according to the embodiment, the CPU 1100 of the computer 1000 executes an information processing program loaded onto the RAM 1200, thereby realizing the functions of the control units such as the learning unit 131 to the generation unit 133 shown in Fig. 10. The information processing program according to the present disclosure and data in the storage device are stored in the HDD 1400.

[0183] The CPU 1100 reads and executes the program data 1450 from the HDD 1400. However, as another example, the CPU 1100 can also obtain these programs from other devices via an external network 1550.

[0184] (5. Supplementary Information) The present technology may also be configured as follows. (1) An information processing system comprising: an acquisition unit that acquires a content program for a first platform and an API (Application Programming Interface) header for a second platform different from the first platform; and a generation unit that generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform. (2) The information processing system described in (1), wherein the generation unit generates content for the second platform including the connection program for the second platform. (3) The information processing system described in (2), wherein the generation unit generates, as the content for the second platform, content for a PC corresponding to the first platform converted into content for an HMD corresponding to the second platform. (4) The information processing system described in any one of (1) to (3), wherein the acquisition unit acquires, as the API header for the second platform, an API header for a second platform for which a smaller amount of content is available than for the first platform. (5) The information processing system according to any one of (1) to (4), wherein the generation unit generates a connection program for the second platform based on a pre-trained language model. (6) The information processing system according to (5), wherein the generation unit uses, as the pre-trained language model, a pre-trained language model that receives as input a general-purpose content program and a general-purpose API header. (7) The information processing system according to (6), wherein the generation unit uses, as the pre-trained language model, a language model that has been pre-trained on how to call the general-purpose API header from the general-purpose content program.(8) The information processing system according to (6) or (7), wherein, when the general-purpose content program is input as the pre-trained language model, the generation unit uses a pre-trained language model that receives as input a content program converted into machine language instructions. (9) The information processing system according to any one of (6) to (8), wherein the generation unit uses, as the pre-trained language model, a language model that has been subjected to supervised learning so that a general-purpose connection program input to the language model and a general-purpose connection program output from the language model substantially coincide. (10) The information processing system according to any one of (6) to (9), wherein, as the pre-trained language model, the generation unit uses a pre-trained language model that outputs a connection program related to the second platform. (11) The information processing system according to (10), wherein, when the connection program related to the second platform is output as the pre-trained language model, the generation unit uses a pre-trained language model that outputs a connection program related to the second platform converted into machine language instructions. (12) The information processing system according to (10) or (11), wherein the generation unit uses, as the pre-trained language model, a language model that has been subjected to supervised learning so that a connection program for the second platform input to the language model and a connection program for the second platform output from the language model substantially coincide. (13) The information processing system according to any one of (10) to (12), wherein the generation unit uses, as the pre-trained language model, a pre-trained language model that receives an API header for the second platform as input. (14) The information processing system according to (13), wherein the generation unit uses, as the pre-trained language model, a language model that has been pre-trained with at least one of knowledge regarding a method of calling an API header for the second platform from the general-purpose content program and knowledge regarding a program that connects the general-purpose content program with the API header for the second platform.(15) The information processing system according to (13) or (14), wherein the generation unit uses, as the pre-trained language model, a pre-trained language model that receives as input a content program for the first platform. (16) The information processing system according to (10), wherein the generation unit uses, as the pre-trained language model, a language model that has been reinforced learned based on verification results of content for a second platform, including a connection program for a second platform, output from the pre-trained language model. (17) The information processing system according to (16), wherein the generation unit uses, as the pre-trained language model, a language model whose parameters have been updated by reinforcement learning until it is verified that no link errors occur in the content for the second platform. (18) The information processing system according to (16) or (17), wherein the generation unit uses, as the pre-trained language model, a language model whose parameters have been updated by reinforcement learning until it is verified that no abnormalities occur in the function of the content for the second platform. (19) An information processing method including: by a computer acquiring a content program for a first platform and an API header for a second platform different from the first platform, and generating a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform. (20) An information processing program for causing a computer to function as an information processing system, comprising: an acquisition unit that acquires the content program for the first platform and an API header for the second platform different from the first platform, and a generation unit that generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform.

[0185] REFERENCE SIGNS LIST 100 Information processing device 110 Communication unit 120 Storage unit 130 Control unit 131 Learning unit 132 Acquisition unit 133 Generation unit

Claims

1. An information processing system comprising: an acquisition unit that acquires a content program for a first platform and an API (Application Programming Interface) header for a second platform different from the first platform; and a generation unit that generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform.

2. The information processing system according to claim 1, wherein the generation unit generates content for the second platform including a connection program for the second platform.

3. The information processing system according to claim 2, wherein the generation unit generates content for the second platform that is converted from content for a PC (Personal Computer) corresponding to the first platform into content for an HMD (Head Mount Display) corresponding to the second platform.

4. The information processing system according to claim 1, wherein the acquisition unit acquires, as the API header related to the second platform, an API header related to a second platform that has a smaller number of available contents than the first platform.

5. The information processing system according to claim 1, wherein the generation unit generates a connection program for the second platform based on a pre-trained language model.

6. The information processing system according to claim 5, wherein the generation unit uses a pre-trained language model that receives a general-purpose content program and a general-purpose API header as input as the pre-trained language model.

7. The information processing system according to claim 6, wherein the generation unit uses, as the pre-trained language model, a language model that has been pre-trained on how to call the general-purpose API header from the general-purpose content program.

8. The information processing system of claim 6, wherein when the general-purpose content program is input, the generation unit uses a pre-trained language model that receives as input a content program converted into machine language instructions.

9. The information processing system according to claim 6, wherein the generation unit uses, as the pre-trained language model, a language model that has been subjected to supervised learning so that a general-purpose connection program input to the language model and a general-purpose connection program output from the language model approximately match.

10. The information processing system according to claim 6, wherein the generation unit uses a pre-trained language model that outputs a connection program related to the second platform as the pre-trained language model.

11. The information processing system described in claim 10, wherein when a connection program for the second platform is to be output, the generation unit uses a pre-trained language model that outputs a connection program for the second platform converted into machine language instructions as the pre-trained language model.

12. The information processing system described in claim 10, wherein the generation unit uses, as the pre-trained language model, a language model that has been supervised and trained so that a connection program for a second platform input to the language model and a connection program for a second platform output from the language model are approximately identical.

13. The information processing system according to claim 10, wherein the generation unit uses a pre-trained language model that receives an API header related to the second platform as input as the pre-trained language model.

14. The information processing system described in claim 13, wherein the generation unit uses, as the pre-trained language model, a language model that has been pre-trained with at least one of knowledge regarding how to call an API header related to a second platform from the general-purpose content program and knowledge regarding a program that connects the general-purpose content program with an API header related to the second platform.

15. The information processing system according to claim 13, wherein the generation unit uses a pre-trained language model that uses a content program for the first platform as input as the pre-trained language model.

16. The information processing system according to claim 10, wherein the generation unit uses, as the pre-trained language model, a language model that has been reinforced learned based on verification results of content for the second platform, including a connection program for the second platform, output from the language model.

17. The information processing system according to claim 16, wherein the generation unit uses, as the pre-trained language model, a language model whose parameters have been updated by the reinforcement learning until it is verified that no link errors occur in the content for the second platform.

18. The information processing system described in claim 16, wherein the generation unit uses, as the pre-trained language model, a language model whose parameters have been updated by the reinforcement learning until it is verified that no abnormalities occur in the functionality of the content for the second platform.

19. An information processing method including: a computer acquiring a content program for a first platform and an API header for a second platform different from the first platform; and generating a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform.

20. An information processing program for causing a computer to function as an information processing system, comprising: an acquisition unit that acquires a content program for a first platform and an API header for a second platform different from the first platform; and a generation unit that generates a connection program for the second platform based on the acquired content program for the first platform and the API header for the second platform.

Images