Estimation device, estimation method, and program

Abstract

One aspect of the present invention is an estimation device comprising a control unit that executes a first mathematical model that is a trained mathematical model for estimating, on the basis of an input word or sentence, word or sentence candidates by adding information to the input word or sentence, wherein the control unit outputs a candidate, which satisfies a predetermined condition among the candidates estimated by the first mathematical model, to an output destination candidate that is a predetermined output destination as an input to a predetermined diffusion model.

Description

Estimation device, estimation method, and program

[0001] The present invention relates to an estimation device, an estimation method, and a program.

[0002] In recent years, there has been growing interest in diffusion models, a type of generative artificial intelligence (AI). Diffusion models accept words or sentences as input and generate data such as image data or audio data in response to the input.

[0003] Ho, Jonathan, Ajay Jain, and Pieter Abbeel. “Denoising diffusion probabilistic models.” Advances in Neural Information Processing Systems 33 (2020): 6840-6851. Radford, Alec, et al. “Learning transferable visual models from natural language supervision.” International conference on machine learning. PMLR, 2021.

[0004] The input to a diffusion model can be any word or sentence. In other words, a diffusion model has few restrictions on the input. However, the data output by the diffusion model changes depending on the input word or sentence. Therefore, the user needs to input the appropriate word or sentence to obtain the desired output.

[0005] However, people cannot always recall the appropriate words or phrases, and there are many situations where they have an image in their head but cannot verbalize it. As a result, there are cases where the burden on users is great when it comes to recalling words or phrases to input into the diffusion model.

[0006] In view of the above circumstances, an object of the present invention is to provide a technology that reduces the burden on users who use diffusion models.

[0007] One aspect of the present invention is an estimation device that includes a control unit that executes a first mathematical model, which is a pre-trained mathematical model that estimates candidate words or sentences with information added to the input words or sentences based on the input words or sentences, and the control unit outputs candidates that satisfy predetermined conditions from the candidates estimated by the first mathematical model to a candidate output destination that is a predetermined output destination as input to a predetermined diffusion model.

[0008] One aspect of the present invention is an estimation method including a control step of executing a first mathematical model, which is a pre-trained mathematical model that estimates word or sentence candidates with information added to the input word or sentence based on the input word or sentence, and in the control step, candidates estimated by the first mathematical model that satisfy predetermined conditions are output to a candidate output destination that is a predetermined output destination as input to a predetermined diffusion model.

[0009] One aspect of the present invention is a program for causing a computer to function as the above-described estimation device.

[0010] The present invention makes it possible to reduce the burden on users who use diffusion models.

[0011] While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments.

[0012] 1 is an explanatory diagram illustrating an estimation apparatus 1 according to an embodiment. The estimation apparatus 1 includes a control unit 11 having a processor 91 such as a CPU (Central Processing Unit) and a memory 92 connected via a bus, and executes a program.

[0013] The control unit 11 executes a first mathematical model, which is a pre-trained mathematical model that, based on an input word or sentence, estimates candidate words or sentences to which information is added for the input word or sentence.

[0014] For example, if the input word or sentence is "red car," the candidate estimated by the first mathematical model is "cool red car." In this case, the added information is "cool." For example, if the input word or sentence is "white dog," the candidate estimated by the first mathematical model is "cute white dog." In this case, the added information is "cute."

[0015] The control unit 11 outputs candidates that satisfy predetermined conditions (hereinafter referred to as "determination conditions") from among the candidates estimated by the first mathematical model to a candidate output destination that is a predetermined output destination as an input to a predetermined diffusion model. The candidate output destination may be a user output destination that is a predetermined output destination whose output can be perceived by a user, or may be the predetermined diffusion model itself. The user output destination may be, for example, a display device. The user output destination may also be, for example, a speaker. The candidate output destination outputs the candidates that have been output by the control unit 11 as input to the diffusion model.

[0016] <Effects of the Estimation Device 1> When a candidate output destination is a user output destination, the words or sentences to be input to the diffusion model are generated by the first mathematical model. Therefore, the user only needs to input the generated words or sentences to the diffusion model. This reduces the number of words or sentences that the user needs to recall. Therefore, the estimation device 1 can reduce the burden on the user who uses the diffusion model.

[0017] If the candidate output destination is a predetermined diffusion model itself, the user can further reduce the effort of inputting the words or sentences generated by the first mathematical model into the diffusion model. Therefore, the estimation device 1 can further reduce the burden on the user who uses the diffusion model.

[0018] <Regarding the Determination Condition> The determination condition is, for example, that the candidate is the one with the highest estimation accuracy among the candidates. The determination condition may also be, for example, that the user instructs that the candidate be input to the diffusion model.

[0019] For example, when the control unit 11 outputs one of the candidates estimated by the first mathematical model to a user output destination, the control unit 11 may output information to the user output destination prompting the user to determine whether or not to input the candidate to the diffusion model. Furthermore, when the user instructs not to input the candidate output by the user output destination to the diffusion model, the control unit 11 may output information to the user output destination prompting the user to input a new word or sentence.

[0020] The conditions instructed by the user to be input to the diffusion model may be, for example, the following detailed conditions.

[0021] The detailed condition is a condition that when the control unit 11 outputs one of the candidates estimated by the first mathematical model to a user output destination, and when the user inputs information to a predetermined input destination based on the information output to the user output destination, the result of execution of the second mathematical model by the control unit 11 is an input instruction result. The second mathematical model is a pre-trained mathematical model that determines, based on the information input to the predetermined input destination, whether the information input to the predetermined input destination is an instruction to input the candidate output to the user output destination to the diffusion model. The input instruction result is a result that is an instruction to input the candidate output to the user output destination to the diffusion model.

[0022] 2 is a diagram illustrating an example of the hardware configuration of the estimation device 1 according to the embodiment. As described above, the estimation device 1 includes the control unit 11 and executes a program. By executing the program, the estimation device 1 functions as a device including the control unit 11, an interface unit 12, and a storage unit 13.

[0023] More specifically, the processor 91 reads out a program stored in the storage unit 13 and stores the read out program in the memory 92. The processor 91 executes the program stored in the memory 92, whereby the estimation device 1 functions as a device including the control unit 11, the interface unit 12, and the storage unit 13.

[0024] The control unit 11 executes, for example, a first mathematical model. The control unit 11 may execute, for example, a second mathematical model. The control unit 11 controls, for example, the operation of each functional unit provided in the estimation device 1. The control unit 11 acquires a user's instruction, for example, via the interface unit 12. The control unit 11 acquires a word or sentence to be input to the first mathematical model, for example, via the interface unit 12. The control unit 11 acquires, for example, information stored in the memory unit 13. The process of acquiring information stored in the memory unit 13 is specifically reading.

[0025] The interface unit 12 includes a communication interface for connecting the estimation device 1 to an external device. The interface unit 12 communicates with the external device via wired or wireless communication. The external device is, for example, a device that transmits a user instruction. In such a case, the interface unit 12 acquires the user instruction by communicating with the device that transmits the user instruction. The external device is, for example, a device that transmits a word or sentence to be input to the first mathematical model. In such a case, the interface unit 12 acquires the word or sentence to be input to the first mathematical model by communicating with the device that transmits the word or sentence to be input to the first mathematical model. The communication interface that constitutes the interface unit 12 is an example of the above-mentioned predetermined input destination.

[0026] The interface unit 12 includes input devices such as a mouse, keyboard, and touch panel. The interface unit 12 may be configured as an interface that connects these input devices to the estimation device 1. In this way, the interface unit 12 accepts input of various information to the estimation device 1 via the input device, either wired or wirelessly. Note that user instructions and words or sentences to be input to the first mathematical model do not necessarily need to be input to a communication interface, but may also be input to an input device. The input devices that constitute the interface unit 12 and the interface that connects the input devices to the estimation device 1 are examples of the above-mentioned predetermined input destinations.

[0027] The interface unit 12 outputs various types of information. The interface unit 12 includes a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display. The interface unit 12 may be configured as an interface that connects these display devices to the estimation device 1. The interface unit 12 outputs, for example, information input to the interface unit 12. The display device that constitutes the interface unit 12 and the interface that connects the display device to the estimation device 1 are examples of user output destinations.

[0028] The storage unit 13 is configured using a computer-readable storage medium device (non-transitory computer-readable recording medium) such as a magnetic hard disk device or a semiconductor storage device. The storage unit 13 stores various information related to the estimation device 1. The storage unit 13 stores, for example, information necessary for executing the first mathematical model. The storage unit 13 stores, for example, information necessary for executing the second mathematical model. The storage unit 13 stores, for example, various information generated by the operation of the control unit 11. The storage unit 13 stores, for example, information acquired by the interface unit 12.

[0029] 3 is a flowchart illustrating an example of a flow of processing executed by the estimation device 1 according to the embodiment. The control unit 11 executes a first mathematical model (step S101). Next, the control unit 11 outputs candidates that satisfy a determination condition to candidate output destinations (step S102).

[0030] The estimation device 1 configured in this way estimates words or sentences to which information is added by executing the first mathematical model, which reduces the burden on users who use the diffusion model, as described above in <Effects of the estimation device 1>.

[0031] (Modification) The first mathematical model is, for example, ChatGPT.

[0032] The control unit 11 may also output, to the candidate output destination, information indicating that the information to be added to the input word or sentence is less accurate than the input word or sentence itself. In such a case, the diffusion model can make an estimation by distinguishing between the estimation by the first mathematical model and the words recalled by the user.

[0033] Since the estimation of the first mathematical model is merely an estimation of the first mathematical model, it is highly likely to differ from what the user desires compared to the words or sentences input by the user. Therefore, by outputting to the candidate output destinations information indicating that the information added to the input words or sentences is less accurate than the input words or sentences themselves, the diffusion model can estimate a result that is closer to what the user desires.

[0034] The information indicating a low degree of certainty is, for example, information indicating a weight for each word, and is information indicating a small weight for a word with a low degree of certainty.

[0035] The estimation device 1 may be implemented using a plurality of information processing devices connected to each other via a network so as to be able to communicate with each other. In this case, the processes executed by the control unit 11 may be distributed among the plurality of information processing devices.

[0036] Note that all or part of the functions of the estimation device 1 may be realized using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. The program may be transmitted via a telecommunications line.

[0037] Although an embodiment of the present invention has been described above in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs within the scope of the gist of the present invention.

[0038] REFERENCE SIGNS LIST 1... estimation device, 11... control unit, 12... interface unit, 13... storage unit, 91... processor, 92... memory

Claims

1. An estimation device comprising: a control unit that executes a first mathematical model, which is a pre-trained mathematical model that estimates candidates for words or sentences to which information is added based on an input word or sentence, and the control unit outputs candidates that satisfy predetermined conditions from among the candidates estimated by the first mathematical model to a candidate output destination that is a predetermined output destination as an input to a predetermined diffusion model.

2. The estimation device according to claim 1, wherein the predetermined condition is that the candidate with the highest estimation accuracy among the candidates.

3. The estimation device according to claim 1, wherein the predetermined condition is a condition instructed by a user to be input into the diffusion model.

4. The estimation device according to claim 3, wherein the predetermined condition is a condition that, when the control unit causes one of the candidates estimated by the first mathematical model to be output to a user output destination, which is a predetermined output destination perceivable by a user, and when the user inputs information to a predetermined input destination based on the information output to the user output destination, the result of the control unit executing a pre-trained mathematical model that determines whether the information input to the input destination is an instruction to input the candidate output to the user output destination to the diffusion model is an instruction to input the candidate output to the user output destination to the diffusion model.

5. The estimation device described in claim 1, wherein the control unit also outputs to the candidate output destination information indicating that the information to be added to the input word or sentence is less certain than the input word or sentence itself.

6. The estimation device according to claim 1, wherein the candidate output is the diffusion model itself.

7. An estimation method comprising: a control step of executing a first mathematical model, which is a pre-trained mathematical model that estimates candidates for words or sentences to which information has been added based on an input word or sentence; wherein in the control step, candidates estimated by the first mathematical model that satisfy predetermined conditions are output to a candidate output destination, which is a predetermined output destination, as an input to a predetermined diffusion model.

8. A program for causing a computer to function as the estimation device according to any one of claims 1 to 6.

Images