Specification-Compatible Data Estimation Device, Machine Learning Method, Specification-Compatible Data Estimation Method, and Program

Abstract

The purpose of the present disclosure is to perform appropriate machine learning according to various specifications.　For this purpose, the disclosed specification-compliant data estimation device 3 trains a machine learning model through machine learning according to prescribed specifications during the learning phase, and comprises: an acquisition unit 31 that acquires data of a plurality of training examples, which are pairs of input data and output data, and a specification function that outputs whether or not each of the plurality of training examples satisfies the prescribed specifications; a specification verification unit 32 that uses the specification function to verify data of a prescribed training example that satisfies the prescribed specifications, among the data of the plurality of training examples; and a machine learning unit 35 that trains the machine learning model through machine learning using the data of the prescribed training example.

Description

Technical Field

[0001] The present disclosure relates to a technique that enables input / output data to satisfy a predetermined specification and reduce a prediction error when there is a predetermined specification that the pair of input / output data for any machine learning model should satisfy.

Background Art

[0002] Let X be an input domain and Y be an output domain. Machine learning is a sequence of pairs of an element x of X i and an element y of Y i such that, when a sequence S = [(x1, y1), (x2, y2),..., (x n , y n )] of training examples is given, a hypothesis h: X → Y that is a mapping from the input domain X to the output domain Y is selected from a set H of hypotheses that can predict y i corresponding to x i as correctly as possible. That is, machine learning is an algorithm for selecting a predetermined hypothesis from a given set H of hypotheses such that there are many training examples (x i , y i ) ∈ S that satisfy h(x i , y i ) = y

[0003] For example, as the input domain X, a set of N-dimensional real vectors, a set of images, a set of news articles, etc. can be considered. Also, as each output domain Y for each input domain X, a set of real numbers, a set of names of objects shown in an image, a set of news categories, etc. can be considered. In particular, when the output domain Y is a finite set Y = {1, 2,..., K} of discrete natural numbers , the hypothesis h ∈ H is called a multi-class classifier. Also, the problem of estimating the corresponding output domain y ∈ Y using a hypothesis for a given input domain x ∈ X is called a multi-class classification problem.

[0004] Methods for solving multi-class classification problems using such machine learning techniques are widely used. For example, the document classification problem is a problem of estimating the genre of a news document given as input, and this problem can be regarded as a multi-class classification problem and solved using machine learning techniques.

[0005] Machine learning techniques are used in a wide range of application fields. However, depending on the scenario where machine learning techniques are used, for a given input domain x, the value that the output domain y = h(x) of the hypothesis should satisfy may be predetermined by the specifications. For example, a mechanism for recommending products that a user is expected to be interested in to users of an online shopping service can be formulated as a multi-class classification problem with the set of users as X and the set of products to be recommended as Y. When solving such a problem using machine learning techniques, the hypothesis h: X → Y estimated from the training (learning) examples may recommend any product to any user.

[0006] However, if it is prohibited by the service agreement or the like to recommend a specific group of products to minors, then for a minor user x, the product y = h(x) output by the hypothesis may be in violation of the agreement if recommended.

[0007] Usually, it is difficult to guarantee that the hypothesis h ∈ H estimated from the set of hypotheses H satisfies the specifications (agreements). Therefore, conventionally, when such specifications exist, as a machine learning model (set of hypotheses) that can take the specifications into account, for example, a specific model such as Markov Logic Networks had to be used (see Non-Patent Document 1).

Prior Art Documents

Non-Patent Documents

[0008]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0009] However, there are various specifications not limited to the above example of minors, and machine learning models that exhibit good performance according to different specifications are different. Therefore, in order to ensure meeting the specifications, a specific machine learning model corresponding to the specification content must be selected. However, it is actually difficult to select an appropriate specific model from various machine learning models.

[0010] The present invention has been made in view of the above points, and aims to perform appropriate machine learning according to various specifications.

Means for Solving the Problems

[0011] To solve the above problems, the invention according to claim 1 is a specification-corresponding data estimation device that machine-learns a machine learning model corresponding to a predetermined specification in a learning phase, and includes an acquisition unit that acquires data of a plurality of training examples that are pairs of input data and output data, and a specification function that outputs whether each of the plurality of training examples satisfies the predetermined specification; a specification verification unit that verifies data of predetermined training examples that satisfy the predetermined specification among the data of the plurality of training examples using the specification function; and a machine learning unit that machine-learns the machine learning model with the data of the predetermined training examples.

Effects of the Invention

[0012] As described above, according to the present invention, there is an effect that appropriate machine learning can be performed according to various specifications.

Brief Description of the Drawings

[0013]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Embodiments for Carrying Out the Invention

[0014] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0015] 〔System Configuration of the Embodiment〕 First, with reference to FIG. 1, an overall configuration outline of the communication system of this embodiment will be described. FIG. 1 is an overall configuration diagram of the communication system according to this embodiment.

[0016] As shown in FIG. 1, the communication system 1 of this embodiment is constructed by a specification-corresponding data estimation device 3 and a communication terminal 5. The communication terminal 5 is managed and used by a user. The user is a person who refers to the output result of the specification-corresponding data estimation device and then determines the subsequent actions.

[0017] Also, the specification-corresponding data estimation device 3 and the communication terminal 5 can communicate via a communication network 100 such as the Internet. The connection form of the communication network 100 can be either wireless or wired.

[0018] The specification-compliant data estimation device 3 is composed of one or more computers. When the specification-compliant data estimation device 3 is composed of a plurality of computers, it may be referred to as the "specification-compliant data estimation device" or the "specification-compliant data estimation system".

[0019] The specification-compliant data estimation device 3 performs appropriate machine learning and estimation according to various specifications (regulations) such as the prohibition of recommending a specific product group to minors.

[0020] The communication terminal 5 is a computer. In FIG. 1, as an example, a notebook personal computer is shown. In FIG. 1, the user operates the communication terminal 5. Note that the processing may be performed by the specification-compliant data estimation device 3 alone without using the communication terminal 5.

[0021] 〔Hardware Configuration〕 <Hardware Configuration of the Specification-Compliant Data Estimation Device> Next, with reference to FIG. 2, the electrical hardware configuration of the specification-compliant data estimation device 3 will be described. FIG. 2 is an electrical hardware configuration diagram of the specification-compliant data estimation device.

[0022] As a computer, the specification-compliant data estimation device 3 includes a CPU (Central Processing Unit) 301, a ROM (Read Only Memory) 302, a RAM (Random Access Memory) 303, an SSD (Solid State Drive) 304, an external device connection I / F (Interface) 305, a network I / F 306, a media I / F 309, and a bus line 31 0, as shown in FIG. 2.

[0023] Among these, the CPU 301 controls the operation of the entire specification-compliant data estimation device 3. The ROM 302 stores programs used for driving the CPU 301 such as the IPL (Initial Program Loader). The RAM 303 is used as the work area of the CPU 301.

[0024] The SSD 304 reads or writes various data according to the control of the CPU 301. Note that an HDD (Hard Disk Drive) may be used instead of the SDD 304.

[0025] The external device connection I / F 305 is an interface for connecting various external devices. External devices in this case include a display, a speaker, a keyboard, a mouse, a USB (Universal Serial Bus) memory, and a printer, etc.

[0026] The network I / F 306 is an interface for data communication via the communication network 100.

[0027] The media I / F 309 controls the reading or writing (storage) of data to / from a recording medium 309m such as a flash memory. The recording medium 309m includes DVDs (Digital Versatile Discs) and Blu-ray Discs (registered trademarks), etc.

[0028] The bus line 310 is an address bus, a data bus, etc. for electrically connecting each component such as the CPU 301 shown in FIG. 2.

[0029] <Hardware Configuration of Communication Terminal> Next, the electrical hardware configuration of the communication terminal 5 will be described with reference to FIG. 3. FIG. 3 is an electrical hardware configuration diagram of the communication terminal.

[0030] As a computer, the communication terminal 5 includes a CPU 501, a ROM 502, a RAM 503, an SSD 504, an external device connection I / F (Interface) 505, a network I / F 506, a display 507, a pointing device 508, a media I / F 509, and a bus line 510 as shown in FIG. 3.

[0031] Among these, the CPU 501 controls the operation of the entire communication terminal 5. The ROM 502 stores programs used for driving the CPU 501 such as IPL. The RAM 503 is used as a work area for the CPU 501.

[0032] The SSD 504 reads or writes various data according to the control of the CPU 501. Note that an HDD (Hard Disk Drive) may be used instead of the SSD 504.

[0033] The external device connection I / F 505 is an interface for connecting various external devices. External devices in this case include a display, a speaker, a keyboard, a mouse, a USB memory, and a printer, etc.

[0034] The network I / F 506 is an interface for data communication via the communication network 100.

[0035] The display 507 is a type of display means such as a liquid crystal or an organic EL (Electro Luminescence) that displays various images.

[0036] The pointing device 508 is a type of input means for selecting and executing various instructions, selecting a processing target, moving a cursor, etc. Note that when the user uses a keyboard, the function of the pointing device 508 may be turned off.

[0037] The media I / F 509 controls the reading or writing (storage) of data to / from a recording medium 509m such as a flash memory. The recording medium 509m includes DVDs, Blu-ray Discs (registered trademark), and the like.

[0038] The bus line 510 is an address bus, a data bus, or the like for electrically connecting each component such as the CPU 501 shown in FIG. 4.

[0039] [Functional Configuration of Specification-Compatible Data Estimation Device] Next, the functional configuration of the specification-compatible data estimation device 3 according to the present embodiment in the learning phase and the estimation (prediction) phase will be described.

[0040] Here, let the input domain be X and the output label be Y. Hereinafter, Y is a set of discrete values Y = {1, 2,..., K}. The hypothesis function h is a function that receives an element of X and outputs an element of Y, and is expressed as h as shown.

[0041] [Functional Configuration in Learning Phase] In the learning phase, the specification-compatible data estimation device 3 is based on a set of training examples S' = [(x1, y1), (x2, y2),..., (x n , y n )] and a finite or infinite set H of hypotheses, and selects a hypothesis

[0042] [Equation] that minimizes

[0043] [Equation] Here, as the set of hypotheses H, for example, parameters

[0044] [Equation] This applies to, for example, the set of all neural networks characterized by

[0045]

Number

[0046]

Number

[0047] The loss function is a function that outputs a large value according to the degree of difference between the two inputs. For example, the 0 - 1 loss

[0048]

Number

[0049]

Number

[0050]

Number

[0051] Subsequently, each function of the specification - compliant data estimation device 3 in the learning phase will be described in detail. FIG. 4 is a functional configuration diagram of the specification - compliant data estimation device in the learning phase.

[0052] As shown in FIG. 4, the specification - compliant data estimation device 3 includes an acquisition unit 31, a specification verification unit 32, and a machine learning unit 35. These units are functions realized by instructions from the CPU 301 in FIG. 2 based on a program. Also, a machine learning model 30a is stored in the RAM 303 or the SSD 304.

[0053] The acquisition unit 31 acquires the training example data S (input data x, output data y) and the specification function c from the communication terminal 7. The acquisition unit 31 also serves as an input unit for inputting learning data for machine learning. Specifically, the acquisition unit 31 acquires, from the outside such as the communication terminal 5, the data S = [(x1, y1),..., (x n , y n )] of a plurality of training examples and the specification function c that outputs whether or not each of the plurality of training examples satisfies a predetermined specification A. Here, the specification function c is a function c(x, y) that receives a pair (x, y) of input / output data ∈ X × Y and outputs result information indicating whether or not it satisfies a predetermined specification. The specification function c(x, y) outputs c(x, y) = 1 when the pair (x, y) of input / output data satisfies the specification, and outputs c(x, y) = 0 when it does not. Also, the predetermined specification A includes, for example, an age limit for the provided person, such as when it is prohibited to recommend a specific product group or service group to minors.

[0054] The specification verification unit 32 verifies the data S' of a predetermined training example that satisfies the predetermined specification A among the data S of the plurality of training examples by using the specification function c acquired by the acquisition unit 31. In this case, the specification verification unit 32 manages (holds) the specification function c for use in the estimation phase. Specifically, the specification verification unit 32 uses the specification function c acquired from the acquisition unit 31 to verify whether or not the pair (x , y i , y i ) in the training example S satisfies the above-mentioned predetermined specification A.

[0055] The machine learning unit 35 performs machine learning on the machine learning model 30g based on the hypothesis h (input data x, output data y) that minimizes the training loss with respect to the set S' of training examples. Specifically, the machine learning unit 35 receives only the predetermined training example S' that satisfies the predetermined specification A from the specification verification unit 32 and determines the hypothesis that minimizes the training loss.

[0056]

Number

[0057] <Functional Configuration of the Estimation Phase> The specification - compliant data estimation device 3 in the inference phase, based on the hypothesis estimated in the learning phase

[0058]

Number

[0059]

Number

[0060]

Number

[0061]

Number

[0062] Subsequently, each function of the specification - compliant data estimation device 3 in the estimation phase will be described in detail. Figure 5 is a functional configuration diagram of the specification - compliant data estimation device in the estimation phase.

[0063] As shown in Fig. 5, the specification correspondence data estimation device 3 has an acquisition unit 31, a specification verification unit 32, an estimation unit 37, and an output unit 39. These units are functions realized by instructions from the CPU 301 in Fig. 2 based on a program. In addition, the RAM 303 or the SSD 304 stores the trained machine learning model 30b. Note that the same reference numerals are used for the functional configurations similar to those in the learning phase, and the description thereof will be omitted.

[0064] The estimation unit 37 uses the trained machine learning model 30b and estimates and outputs output data y that satisfies a predetermined specification by a specification function c managed by the specification verification unit 32. Specifically, the estimation unit 37 uses the hypothesis obtained in the learning phase

[0065]

number

[0066] In addition, y is

[0067]

number

[0068] The output unit 39 indicates the output data y that satisfies the predetermined specifications estimated by the estimation unit 37. Information on the estimation result is output from the specification correspondence data estimation device 3. Examples of output include displaying the information on a display connected to the external device connection I / F 305 in Fig. 2, transmitting the information to an external device via the network I / F 306, etc.

[0069] [Processing or operation of the specification-compliant data estimation device] Next, the processing or operation of the specification correspondence data estimation device 3 in the learning phase and the estimation phase will be described with reference to FIGS.

[0070] <Processing or Actions in the Learning Phase> FIG. 6 is a flowchart showing the processes or operations executed by the specification-corresponding data estimation device in the learning phase.

[0071] S11: The acquisition unit 31 acquires a training example S (input data x, output data y) and a specification function c based on an input from the communication terminal 7 or a direct input to the own device (specification-corresponding data estimation device).

[0072] S12: The specification verification unit 32 outputs a set S' of predetermined training examples that satisfy the specification function c acquired by the acquisition unit 31 from the training example S acquired by the acquisition unit 31. In this case, the specification verification unit 32 holds the specification function c for use in the estimation phase. For example, the specification verification unit 32 selects all samples that satisfy the specification function c(x

[0073] ,y i , i ) = 1 from the training example data S input from the acquisition unit 31 to create the set data

[0074]

Number

[0075] S13: The machine learning unit 35 trains the machine learning model 30a based on a hypothesis h (input data x, output data y) that minimizes the training loss with respect to the set data S' of training examples. The machine learning model 30a uses an existing algorithm.

[0076] This theoretically shows that the specification-corresponding data estimation device 3 can select a hypothesis with less estimation error while taking into account external specifications.

[0077] The description of the processes or operations in the learning phase ends as above.

[0078] <Processes or operations in the estimation phase> FIG. 7 is a flowchart showing processing or operations executed by the specification-compliant data estimation device in the estimation phase.

[0079] S21: The acquisition unit 31 acquires input data x ∈ X based on input from the communication terminal 7 or direct input to the own device (specification-compliant data estimation device 3).

[0080] S22: The estimation unit 37 uses the learned machine learning model 30b and outputs output data y that satisfies a predetermined specification A by means of the specification function c managed (held) by the specification verification unit 32. to do.

[0081] S23: The output unit 39 outputs information on the estimation result.

[0082] Thus, the description of the processing or operations in the estimation phase is completed.

[0083] 〔Effects of the Embodiment〕 As described above, according to the present embodiment, there is an effect that appropriate machine learning can be performed according to various specifications (conventions).

[0084] 〔Supplementary Explanation〕 The present invention is not limited to the above-described embodiment, and may have the following configurations or processes (operations). (1) The specification-compliant data estimation device 3 can also be realized by a computer and a program, but it is also possible to record this program on a (non-transitory) recording medium or provide it via the communication network 100. (2) In the communication between the specification-compliant data estimation device 3 and the communication terminal 5, another device (such as a server or a router) may relay the data. For example, in this specification, for the sake of simplicity, it is described that the input unit 31 of the specification-compliant data estimation device 3 transmits data to the communication terminal 5, but this transmission process also includes the case where another device relays the data. (3) In the above embodiment, a notebook personal computer is shown as an example of the communication terminal 5, but the present invention is not limited thereto. For example, a desktop personal computer, a tablet terminal, a smartphone, a smartwatch, a car navigation device, a refrigerator, a microwave oven, etc. may also be used. (4) Each of the CPUs 301 and 501 may be not only single but also plural.

Description of Reference Numerals

[0085] 1 Communication system 3 Specification-Compatible Data Estimation Device 5 Communication terminal 30a Machine learning model 30b Trained machine learning model 31 Acquisition unit (input unit) 32 Specification verification unit 35 Machine learning unit 37 Estimation unit 39 Output unit

Claims

1. A specification-corresponding data estimation device that machine-learns a machine learning model to correspond to a predetermined specification in a learning phase, comprising: an acquisition unit that acquires data of a plurality of training examples that are pairs of input data and output data, and a specification function that outputs whether or not each of the plurality of training examples satisfies the predetermined specification; a specification verification unit that verifies data of a predetermined training example that satisfies the predetermined specification among the data of the plurality of training examples using the specification function; a machine learning unit that machine-learns the machine learning model using the data of the predetermined training example; A specification-corresponding data estimation device having the above.

2. The specification-corresponding data estimation device according to claim 1, wherein the predetermined specification indicates an age limit for a person to be provided.

3. A machine learning method executed by a specification-corresponding data estimation device that machine-learns a machine learning model to correspond to a predetermined specification in a learning phase, comprising: the specification-corresponding data estimation device an acquisition process of acquiring data of a plurality of training examples that are pairs of input data and output data, and a specification function that outputs whether or not each of the plurality of training examples satisfies the predetermined specification; a specification verification process of verifying data of a predetermined training example that satisfies the predetermined specification among the data of the plurality of training examples using the specification function; a machine learning process of machine-learning the machine learning model using the data of the predetermined training example; A machine learning method for executing the above.

4. A program for causing a computer to execute the method according to claim 3.

5. A specification-corresponding data estimation device that performs an estimation corresponding to a predetermined specification using a learned machine learning model in an estimation phase, comprising: an acquisition unit that acquires predetermined input data; an estimation unit that uses a learned machine learning model and estimates output data that satisfies the predetermined specification by a specification function that outputs whether or not each of a plurality of training examples that are pairs of input data and output data satisfies the predetermined specification; an output unit that outputs information on an estimation result indicating the estimated output data; A specification-corresponding data estimation device having the above.

6. The specification-corresponding data estimation device according to claim 5, wherein the predetermined specification indicates an age limit for a person to be provided.

7. A specification-corresponding data estimation method executed by a specification-corresponding data estimation device that performs an estimation corresponding to a predetermined specification using a learned machine learning model in an estimation phase, comprising: the specification-corresponding data estimation device An acquisition process for acquiring predetermined input data, An estimation process for estimating output data that satisfies the predetermined specification by using a learned machine learning model and a specification function that outputs whether or not each of a plurality of training examples, which are pairs of input data and output data, satisfies the predetermined specification, An output process for outputting information on an estimation result indicating the estimated output data, A specification-corresponding data estimation method for executing the above. [

8. ] A program for causing a computer to execute the method according to Claim 7.

Images