Machine learning device, machine learning method, and program
Patent Information
- Application Number
- JP2025560476
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Filing Date
- 2023-11-30
- Publication Date
- 2025-06-05
AI Technical Summary
Existing methods, such as Bayesian optimization, struggle to find inflection points in psychometric functions, which are crucial for understanding subjective data like emotions and sensory experiences.
A machine learning device that uses Bayesian optimization to efficiently collect subjective data by applying an evaluation function that maximizes at the inflection point of the psychometric function, allowing for the selection and presentation of optimal parameters to users.
Enables the efficient collection of subjective data and the identification of inflection points, facilitating the construction of models related to subjective experiences and improving personalization in fields like emotion recognition and sensory evaluation.
Abstract
Description
Machine learning device, machine learning method, and program
[0001] One aspect of the present invention relates to a machine learning device, a machine learning method, and a program.
[0002] Personalization refers to optimizing a target based on an individual's attributes, hobbies, and preferences. For example, products, content, services, and advertisements are personalized based on log data such as browsing, playback, clicks, purchases, and ratings. Personalization has become common in fields where it is easy to create a system for collecting personal data.
[0003] On the other hand, personalization is not common in fields where it is difficult to create a system for collecting personal data. For example, personalization of how people feel about something, such as evaluations based on the five human senses, is not common. In other words, data based on the user's subjective opinion (emotions / likability ratings estimated from facial expressions, gestures, and voices), or perceptions of the aroma, taste, and deliciousness of food and drink, are difficult to personalize. If this limitation could be overcome and personalization of subjective data were realized, it could be used to accurately convey emotions by converting the facial expressions (of an avatar) to match the emotion recognition characteristics of a certain individual (person A), or to prepare a dish that person A finds most delicious. Against this background, a method for efficiently collecting subjective data is needed to easily build models related to subjective data.
[0004] Bayesian optimization is known as a method of obtaining an optimal value by collecting a small amount of data (see, for example, Non-Patent Document 1). This is a technique that uses a rating for a certain parameter as input and repeatedly presents the parameters to be searched for. For example, when a user rates a particular recipe, another recipe is presented and the user is prompted to rate it again. By repeating this process, a recipe that maximizes the user's rating can be obtained. Reference [1] discloses an example of optimizing a cookie recipe using Bayesian optimization.
[0005] B. Shahriari, K. Swersky, Z. Wang, R. Adams, and N. de Freitas : Taking the human out of the loop: A Review of Bayesian Optimization. Proc. of the IEEE,(1), 12 / 2015(2016)
[0006] While Bayesian optimization can generally obtain maximum and minimum scores, it cannot obtain the inflection points of a function. Psychometric functions, which are often used in this field, contain inflection points, so some kind of technological innovation is required. A psychometric function is a function that expresses the relationship between the physical quantity of a stimulus and the response to that stimulus. Its inflection point represents, for example, the minimum facial expression intensity that is perceived as a smile.
[0007] The present invention has been made in light of the above circumstances, and aims to provide a technique that makes it possible to search for an inflection point using Bayesian optimization.
[0008] A machine learning device according to one aspect of the present invention includes a memory unit and a processor. The memory unit stores response data including presented parameters related to the measurement of a psychometric function and a user's response to the parameters. The processor includes an evaluation function application unit, a model learning unit, a parameter selection unit, and a parameter presentation unit. The evaluation function application unit applies an evaluation function to the response result to obtain an evaluation value. The model learning unit repeatedly provides the parameters and evaluation value of the response data to a machine learning model for machine learning, thereby generating a trained model corresponding to the psychometric function. The parameter selection unit selects parameters to be presented by Bayesian optimization based on the evaluation value and the trained model. The parameter presentation unit presents the selected parameters to the subject.
[0009] According to one aspect of the present invention, it is possible to search for inflection points using Bayesian optimization.
[0010] FIG. 1 is a functional block diagram showing an example of a machine learning device according to an embodiment. FIG. 2 is a diagram showing an example of data stored in the answer result database 31a. FIG. 3 is a diagram showing an example of the flow of data in the machine learning device 10. FIG. 4 is a graph showing an example of a curve drawn by a psychometric function. FIG. 5 is a graph showing an example of a curve having a maximum value. FIG. 6 is a diagram for explaining an evaluation function according to an embodiment. FIG. 7 is a functional block diagram showing an example of an existing device that performs Bayesian optimization. FIG. 8 is a block diagram showing an example of the hardware configuration of the machine learning device 10.
[0011] Emotions are how people feel about certain events, and are heavily dependent on factors such as individual personality, experience, and memory. Therefore, even when given the same information under the same circumstances, the information received varies significantly from person to person. In particular, in interpersonal communication, the way the other person perceives something varies greatly, and it is possible that the information will be received in a way that the sender did not anticipate. Some people tend to interpret the other person's emotions more negatively than necessary, and may even feel uncomfortable with communication itself. Psychometric functions can be useful in smoothing communication between sender and receiver.
[0012] 1 is a functional block diagram showing an example of a machine learning device according to an embodiment. The machine learning device 10 is a computer and includes an input unit 40, a processor 20, a storage unit 30, and an output unit 50.
[0013] The input unit 40 includes a user response input unit 41. The user response input unit 41 accepts input of responses from the user (subject) to presented parameters (stimuli) related to the measurement of the psychometric function. The presented parameters and the user's responses to these parameters are stored as a set in the storage unit 30.
[0014] The storage unit 30 stores a database 31 and a model 32. The database 31 includes an answer result database 31a. The answer result database 31a is a database that holds answer data including parameters and answer results.
[0015] FIG. 2 is a diagram showing an example of data stored in the answer result database 31a. For example, consider optimizing a cookie recipe based on a user's answers. When a user scores a certain recipe (e.g., the amount and ratio of ingredients), another recipe is presented and the user is prompted to score it again. By repeating this process, a recipe with the highest user rating can be obtained. In such a case, the answer result database 31a in FIG. 2 stores parameters that differ for each recipe and the user's answer results, given on a scale of, for example, 10 points, for each attempt.
[0016] 1, the description will be continued. The model 32 in the storage unit 30 is a trained model generated by the machine learning device 10.
[0017] The processor 20 includes an evaluation function application unit 400, a model learning unit 500, a parameter selection unit 201, and a parameter presentation unit 202 as processing functions according to the embodiment.
[0018] The evaluation function application unit 400 applies a prepared evaluation function to the answer given by the user to obtain an evaluation value.
[0019] The model learning unit 500 repeatedly provides the parameters of the response data and the evaluation values to the machine learning model to perform machine learning, thereby generating a learned model corresponding to the psychometric function.
[0020] The parameter selection unit 201 selects parameters to be presented by Bayesian optimization based on the evaluation values and the trained model. The parameter presentation unit presents the selected parameters to the subject.
[0021] The model learning unit 500 repeatedly provides the parameters of the response data stored in the response result database 31a and the evaluation values obtained by the evaluation function application unit 400 to an untrained neural network for machine learning. Through machine learning, a trained model corresponding to the psychometric function is generated. The generated model is stored in the storage unit 30 as the model 32.
[0022] The parameter selection unit 201 selects parameters to be presented using Bayesian optimization based on the evaluation value obtained by the evaluation function application unit 400 and the trained model 32.
[0023] The parameter presenting unit 202 presents the selected parameters to the subject.
[0024] FIG. 3 is a diagram showing an example of the flow of data in the machine learning device 10. The answer results input from the user answer input unit 41 are stored in the answer result database 31a of the storage unit 30. The evaluation function application unit 400 acquires the answer results and applies the evaluation function to obtain an evaluation value. The model learning unit 500 acquires the evaluation value, generates a trained model 32 through machine learning, and stores it in the storage unit 30. The parameter selection unit 201 acquires the evaluation value and the trained model 32 and selects parameters to be presented to the user. The selected parameters are displayed by the parameter presentation unit 202, for example, on a browser screen of a display (not shown), prompting the user to provide an answer. In response, the answer results are input again, and the model learning unit 500 repeatedly fine-tunes the model 32.
[0025] 4 is a graph showing an example of a curve drawn by a psychometric function. When the parameter value is plotted on the horizontal axis and the evaluation value on the vertical axis, the psychometric function draws a growth curve (S-shaped curve) with an inflection point near the center of the domain.
[0026] Next, the machine learning of the embodiment will be described in detail using mathematical expressions. As mentioned above, with general Bayesian optimization, it is possible to obtain the maximum (or minimum) score as shown in Figure 5, but it is not possible to obtain an inflection point represented by a psychometric function. The inflection point of a psychometric function is, for example, a point corresponding to the minimum facial expression intensity perceived as a smile, and contains the following important information:
[0027] Stimulus threshold (the boundary point at which a sensation occurs or does not occur) Stimulus peak (the point at which an increase in intensity beyond which a sensation does not occur) Discrimination threshold (the minimum value at which a change from a certain stimulus is noticed) Subjective equivalence point (the point at which two stimuli are subjectively perceived as equal) Therefore, in this embodiment, an evaluation value is obtained using an evaluation function that is maximum at the inflection point of the psychometric function. For example, the evaluation value z can be obtained using the evaluation function of equation (1).
[0028]
[0029] In equation (1), x represents the parameter value, and y represents the answer value. The answer value may be either a continuous number (e.g., the degree to which it is perceived as a smile) or a binary value (e.g., whether it is perceived as a smile or not).
[0030] 6A and 6B are diagrams illustrating evaluation functions according to an embodiment. Each graph in FIG. 6 shows an example in which the parameter value at the inflection point in the psychometric function is 0.4. FIG. 6A shows the psychometric function of a target, which has an inflection point. The coordinates of this inflection point cannot be obtained by Bayesian optimization.
[0031] Figure 6(b) is a graph plotting the first and second terms on the right-hand side of the evaluation function of Equation (1). The solid line (rising upward to the right) represents the first term on the right-hand side, and the dashed line (sloping downward to the right) represents the second term on the right-hand side. The evaluation function application unit 400 (Figure 2) multiplies the true psychometric function by this evaluation function, for example, to obtain the curve shown in Figure 6(c), and calculates an evaluation value based on the coordinates of its peak. In other words, the peak of the solid line in Figure 6(a) represents the inflection point of the psychometric function.
[0032] The parameters to be evaluated can be found using, for example, the Upper Confidence Bound, Expected Improvement, or Mutual Information algorithm (see Non-Patent Document 1). This allows the inflection point to be interpreted as the optimal value obtained by Bayesian optimization. In other words, Bayesian optimization makes it possible to obtain the location (parameter value) of the inflection point of the psychometric function.
[0033] To explain in more detail, consider the case where a user is presented with facial expressions x[0 to 1] between the maximum smile intensity of 1 and the minimum smile intensity of 0, and is asked to answer whether they are smiling or not using the binary value (0, 1). If the user answers with 1 for smiling and 0 for not smiling, the following results are obtained, for example:
[0034] Presented smile intensity x = 0.2 and answered "not smiling" (0) Presented smile intensity x = 0.5 and answered "not smiling" (0) Presented smile intensity x = 0.7 and answered "smiling" (1) Presented smile intensity x = 0.9 and answered "smiling" (1) Interpreting these results, we can predict that there is an inflection point in the vicinity of 0.5 and 0.7, as the answers change around this range.
[0035] The evaluation function of formula (1) is applied to each answer to obtain an evaluation value z. (2) shows the evaluation value for each smile intensity.
[0036]
[0037] A detailed interpretation of the evaluation value (2) suggests that the maximum value lies somewhere between 0.2 and 0.7. By converting the answer y into an evaluation value z in this way, z = f(x) becomes a convex function that reaches a maximum at the inflection point, making it possible to search for the inflection point using Bayesian optimization. The evaluation value when the answer is "smiling" (y = 1) at the inflection point (smile intensity x = 0.4) is expressed as (3).
[0038]
[0039] For comparison, Figure 7 is a functional block diagram showing an example of an existing device that performs Bayesian optimization. The existing device trains a model without considering the evaluation value of the answer result, and selects parameters to be presented to the user using Bayesian optimization. As a result, it cannot be said that appropriate parameters were always presented, and it was difficult to efficiently collect subjective data.
[0040] In contrast, in the present embodiment, the subject's responses are converted into evaluation values using an evaluation function that is maximized at the inflection point of the psychometric function, and the model is trained using these evaluation values. This makes it possible to search for the inflection point using Bayesian optimization. This in turn makes it possible to efficiently collect subjective data and easily build a model related to the subjective data.
[0041] The functions of the machine learning device 10 can be realized by installing a program on a computer. For example, the functions of the machine learning device 10 can be implemented by having a computer execute a program provided as packaged software or online software. The computer is not limited to a so-called server computer, and may be a desktop or notebook personal computer. Portable terminals such as smartphones and tablets are also included in the category of computers. Furthermore, computer resources virtualized on the cloud can also function as the machine learning device 10.
[0042] 8 is a block diagram showing an example of the hardware configuration of the machine learning device 10. As shown in FIG. 8, the machine learning device 10 includes a CPU (Central Processing Unit) 20A, a bridge circuit 102, memory 60, a GPU (Graphics Processing Unit) 20B connected to a display 51, a BIOS-ROM 70, storage 80, a USB connector 90, and an input unit 40.
[0043] The storage 80 is a non-volatile storage medium (block device), such as a hard disk drive (HDD) or a solid state drive (SSD). The storage 80 stores basic programs such as an operating system (OS) 62 and device drivers, as well as a program 61 for implementing the functions of the machine learning device 10.
[0044] The memory 60 includes a ROM (Read Only Memory) and a RAM (Random Access Memory). The CPU 20A and the GPU 20B are arithmetic elements related to the processor 20. The CPU 20A mainly controls the machine learning device 10. The GPU 20B mainly performs calculations related to image processing (such as multiply-accumulate operations) at high speed. The CPU 20A executes a BIOS (basic input / output system) stored in the BIOS-ROM 70. The CPU 20A also loads a program 61 from the storage 80 into the memory 60 and executes it. The same is true for the GPU 20B.
[0045] The bridge circuit 102 relays data transmission between the CPU 20A and the GPU 20B and each component. For example, the bridge circuit 102 is interposed between the CPU 20A and the GPU 20B and hardware devices connected to a PCI (Peripheral Component Interconnect) bus or a PCIe bus (not shown), relaying communication between them.
[0046] The USB connector 90 connects to a USB device or the like. For example, the program 61 may be installed in the machine learning device 10 via a USB device. The input unit 40 connects the machine learning device 10 to a communication network such as a local area network (LAN) or a wide area network (WAN).
[0047] The program 61 and various data may be stored in a removable storage medium other than the storage 80, and may be read by the CPU 20A from a disk drive or the like. Alternatively, the program 61 and various data may be stored in another computer connected via a communication network, and may be read by the CPU 20A via the input unit 40.
[0048] In short, this invention is not limited to the above-described embodiments, and various modifications can be made in the implementation stage without departing from the spirit of the invention. Furthermore, the embodiments may be implemented in appropriate combinations, in which case the combined effects can be obtained. Furthermore, the above-described embodiments include various inventions, and various inventions can be extracted by combining selected elements from the disclosed elements. For example, if the problem can be solved and the desired effect can be obtained even if some elements are deleted from all elements shown in the embodiments, the configuration from which these elements are deleted can be extracted as an invention.
[0049] <References> [1] “The makings of a smart cookie”, [online], Google, [searched November 13, 2023], Internet,<URL:https: / / www.blog.google / technology / research / makings-smart-cookie / >
[0050] 10...machine learning device 20...processor 20A...CPU 30...memory unit 31...database 31a...answer result database 32...trained model 40...input unit 41...user answer input unit 50...output unit 51...display 60...memory 61...program 80...storage 90...USB connector 102...bridge circuit 201...parameter selection unit 202...parameter presentation unit 400...evaluation function application unit 500...model learning unit.
Claims
1. In a machine learning device comprising a memory unit and a processor, the memory unit stores response data including parameters presented in relation to the measurement of a psychometric function and the user's response results to the parameters, the processor includes an evaluation function application unit that applies an evaluation function to the response results to obtain an evaluation value, a model learning unit that repeatedly provides the parameters of the response data and the evaluation value to a machine learning model to perform machine learning and generate a learned model corresponding to the psychometric function, a parameter selection unit that selects parameters to be presented by Bayesian optimization based on the evaluation value and the learned model, and a parameter presentation unit that presents the selected parameters to a subject. A machine learning device.
2. The machine learning device according to claim 1, wherein the evaluation function is a function that maximizes at the inflection point of the psychometric function.
3. A machine learning method by a computer comprising a memory unit that stores response data including parameters presented in relation to the measurement of a psychometric function and the user's response results to the parameters, and a processor, wherein the processor includes a process of applying an evaluation function to the response results to obtain an evaluation value, a process of repeatedly providing the parameters of the response data and the evaluation value to a machine learning model to perform machine learning and generate a learned model corresponding to the psychometric function, a process of selecting parameters to be presented by Bayesian optimization based on the evaluation value and the learned model, and a process of presenting the selected parameters to a subject. A machine learning method.
4. A program including instructions for causing a computer to function as the machine learning device according to claim 1 or 2.