Learning system

Abstract

To train a generative model so as to reflect individual characteristics.SOLUTION: A learning system includes: storage means for storing user data acquired from users in multiple layers; selection means for selecting a layer to be used as training data of a generative model, from among the layers; and training means which trains the generative model so as to be personalized for the user using the user data included in the selected layer.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] TECHNICAL FIELD This disclosure relates to the technical field of learning systems. [Background technology]

[0002] In recent years, there has been progress in the development of technologies for forming emotional connections between robots and humans. For example, Patent Document 1 discloses a device that learns human emotional interactions in a multimodal manner using feature information extracted from human information (such as acoustic information collected by a microphone). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-006610 Summary of the Invention [Problem to be solved by the invention]

[0004] AI (Artificial Intelligence) that learns by collecting information from people has the potential to be used in a wide range of applications. However, it is not easy to train AI that appropriately reflects individual characteristics.

[0005] This disclosure has been made in consideration of the above-mentioned problems, and aims to provide a learning system that can collect information from people and learn a generative model that appropriately reflects individual characteristics. [Means for solving the problem]

[0006] A learning system according to one aspect of the present disclosure includes a storage means for storing user data acquired from a user divided into multiple layers, a selection means for selecting a layer from the multiple layers to be used as training data for a generative model, and a learning means for training the generative model using the user data included in the selected layer to personalize the generative model for the user. [Brief explanation of the drawings]

[0007] [Figure 1] 1 is a block diagram showing the configuration of a learning system according to a first embodiment. [Figure 2] 4 is a flowchart showing the flow of a learning operation by the learning system according to the first embodiment. [Figure 3] FIG. 2 is a schematic diagram showing a specific learning example using the learning system according to the first embodiment. [Figure 4] 10 is a flowchart showing the flow of a learning operation by the learning system according to the second embodiment. [Figure 5] 11 is a flowchart showing the flow of a learning operation by a learning system according to a third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] Hereinafter, an embodiment of a learning system will be described with reference to the drawings.

[0009] First Embodiment The first learning system will be described with reference to FIGS.

[0010] (System Configuration) First, the configuration of the learning system according to the first embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the learning system according to the first embodiment.

[0011] In FIG. 1, a learning system 1 according to the first embodiment is configured to include a PGAI (Personal Generative AI) 10 and a processing device 20.

[0012] PGAI10 is a specific example of a "generative model." PGAI10 is a generative model that is trained to be personalized to a user (i.e., trained specifically for an individual user). PGAI10 may be a model consisting of a neural network that is machine-learned using deep learning, for example. PGAI10 is trained using user data acquired from a user. For example, PGAI10 may be trained using voice information acquired from a user. More specifically, PGAI10 may be trained using voice information collected by a smart speaker or the like.

[0013] The PGAI 10 may be, for example, a model installed in a communication robot capable of communicating with a user (e.g., through text or voice). In this case, the PGAI 10 may acquire user data through communication with the user. That is, the PGAI 10 may learn through communication with the user.

[0014] The processing device 20 is a device equipped with a processor and a memory. The processor is configured to be able to execute various types of arithmetic processing. The memory is configured to be able to store various types of information. A computer program may be stored in the memory. The processor, together with the memory storing the computer program, may execute various types of processing to be performed by the learning system 1. For example, when the processor executes the computer program, logical functional blocks for executing the processing to be performed by the learning system 1 may be realized within the processor.

[0015] The processing device 20 may be configured to include an input device and an output device in addition to the above-mentioned processor and memory. Specifically, the processing device 20 may include input devices such as a mouse, a keyboard, a touch panel, a microphone, etc. The processing device 20 may also include output devices such as a display, a speaker, etc.

[0016] The processing device 20 includes a user data storage unit 201, a learning data selection unit 202, and a learning unit 203 as functional blocks for realizing its functions. The user data storage unit 201 is a specific example of "storage means". The learning data selection unit 202 is a specific example of "selection means". The learning unit 203 is a specific example of "learning means". The user data storage unit 201 may be realized by a memory or the like included in the processing device 20. Each of the learning data selection unit 202 and the learning unit 203 may be realized by a processor or the like included in the processing device 20.

[0017] The user data storage unit 201 is configured to be able to store user data acquired from a user. The user data storage unit 201 particularly stores the user data divided into multiple layers. The user data storage unit 201 may store the user data divided into layers for each predetermined period, for example. For example, the user data storage unit 201 may store the user data divided into layers for each year. The user data storage unit 201 may also store the user data divided into layers for each organization the user belongs to. For example, the user data storage unit 201 may store the user data acquired by the user when he / she was an elementary school student, the user data acquired by the user when he / she was a junior high school student, the user data acquired by the user when he / she was a high school student, and the user data acquired by the user when he / she was a university student, each as a separate layer. The user data storage unit 201 may be configured to continuously acquire user data from the user over a long period of time (for example, several years to several decades) and accumulate the user data.

[0018] The training data selection unit 202 is configured to be able to select user data to be used for training the PGAI 10 (i.e., user data to be used as training data) from the user data of the multiple layers stored in the user data storage unit 201. The training data selection unit 202 typically selects some of the layers included in the multiple layers as training data. However, the training data selection unit 202 may also select all of the multiple layers as training data. A specific method for selecting user data will be described in detail in another embodiment described later.

[0019] The learning unit 203 is configured to be able to learn the PGAI 10 using the user data selected by the learning data selection unit 202. That is, the learning unit 203 excludes user data not selected by the learning data selection unit 202 from the user data of multiple layers stored in the user data storage unit 201, and uses the selected user data to learn the PGAI 10. Note that the specific learning algorithm used by the learning unit 203 is not particularly limited, and the learning unit 203 may, for example, appropriately adopt existing technology to learn the PGAI 10.

[0020] (Learning procedure) Next, the flow of the learning operation of the learning system 1 according to the first embodiment (specifically, the operation when learning the PGAI 10) will be described with reference to Fig. 2. Fig. 2 is a flowchart showing the flow of the learning operation by the learning system according to the first embodiment.

[0021] 2, when the operation of the learning system 1 according to the first embodiment starts, the learning data selection unit 202 first selects user data to be used for learning the PGAI from the user data of multiple layers stored in the user data storage unit 201 (step S11). Information about the selected user data is output to the learning unit 203.

[0022] Next, the learning unit 203 reads out the user data selected by the learning data selection unit 202 from the user data storage unit 201 (step S12). Then, the learning unit 203 executes learning of the PGAI 10 using the read user data as learning data (step S13). When the learning 203 is completed, the learning unit 203 saves the learned model (step S14).

[0023] (Study example) Next, a specific example of learning performed by the learning system 1 according to the first embodiment will be described with reference to Fig. 3. Fig. 3 is a schematic diagram showing a specific example of learning performed by the learning system according to the first embodiment.

[0024] As shown in FIG. 3, the learning system 1 according to the first embodiment learns the PGAI 10 using a group of user data stored in layers according to the user's affiliation (i.e., a group of user data stored in the user data storage unit 201).

[0025] In the example shown in Figure 3, user data acquired during junior high school years is not selected from the user data group, and user data acquired during other periods is selected. Therefore, the PGAI 10 is trained using data other than the user data acquired during junior high school years. As a result, the PGAI 10 is generated as a model excluding the junior high school years.

[0026] The learning system 1 according to the first embodiment may change the selected user data to learn different PGAIs 10. That is, the learning system 1 may be configured to generate multiple PGAIs 10 for a single user, each using different user data as learning data. For example, a model excluding the college years may be generated in addition to the model excluding the middle school years shown in FIG. 3.

[0027] (Technical Effects) Next, the technical effects obtained by the learning system 1 according to the first embodiment will be described.

[0028] As described with reference to FIGS. 1 to 3, in the learning system 1 according to the first embodiment, the PGAI 10 is trained using selected user data from among user data stored in multiple layers. This allows the generation of a PGAI 10 that appropriately reflects the characteristics of the individual user. Therefore, for example, the PGAI 10 can be used to optimize self-promotion. The output of the PGAI 10 can be used, for example, to match users with companies.

[0029] Second Embodiment A learning system according to the second embodiment will be described with reference to Figure 4. The second embodiment differs from the first embodiment in some configurations and operations, but other parts may be the same as the first embodiment. Therefore, the following will describe in detail the parts that differ from the first embodiment, and will omit explanations of other overlapping parts as appropriate.

[0030] (Learning procedure) The flow of the learning operation of the learning system 1 according to the second embodiment will be described with reference to Fig. 4. Fig. 4 is a flowchart showing the flow of the learning operation by the learning system according to the second embodiment. Note that in Fig. 4, the same processes as those shown in Fig. 2 are denoted by the same reference numerals.

[0031] As shown in FIG. 4, when the operation of the learning system 1 according to the second embodiment starts, the learning data selection unit 202 first accepts an operation by the user (step S21). The operation here is an operation in which the user selects user data to be used for learning. The user may perform the selection operation using various input devices. At this time, the learning data selection unit 202 may display the layers of the user data stored to the user, allowing the user to select which layer to use for learning.

[0032] The learning data selection unit 202 selects user data to be used for learning the PGAI based on the user's operation (step S22). Information about the selected user is output to the learning unit 203.

[0033] Next, the learning unit 203 reads out the user data selected by the learning data selection unit 202 from the user data storage unit 201 (step S12). Then, the learning unit 203 executes learning of the PGAI 10 using the read user data as learning data (step S13). When the learning 203 is completed, the learning unit 203 saves the learned model (step S14).

[0034] (Technical Effects) Next, the technical effects obtained by the learning system 1 according to the second embodiment will be described.

[0035] As described in FIG. 4, in the learning system 1 according to the second embodiment, user data to be used for learning is selected based on user operation. In this way, it is possible to generate the PGAI 10 while reflecting the user's intention. For example, the user may select from the stored user data a layer that the user determines to accurately reflect his or her characteristics. In this way, the PGAI 10 can be learned more appropriately than when user data is selected without the user's intention.

[0036] Third Embodiment A learning system according to the third embodiment will be described with reference to Figure 5. The third embodiment differs in some configurations and operations from the first and second embodiments described above, but other parts may be similar to the first and second embodiments. Therefore, the following will describe in detail the parts that differ from the embodiments already described, and will omit explanations of other overlapping parts as appropriate.

[0037] (Learning procedure) The flow of the learning operation of the learning system 1 according to the third embodiment will be described with reference to Fig. 5. Fig. 5 is a flowchart showing the flow of the learning operation by the learning system according to the third embodiment. Note that in Fig. 5, the same processes as those shown in Fig. 2 and Fig. 4 are denoted by the same reference numerals.

[0038] As shown in FIG. 5, when the operation of the learning system 1 according to the third embodiment is started, the learning data selection unit 202 first calculates the degree of discrepancy between layers for the user data of multiple layers stored in the user data storage unit 201 (step S31). The learning data selection unit 202 may, for example, calculate the degree of discrepancy by quantifying the features of the user data. More specifically, the learning data selection unit 202 may calculate the degree of discrepancy between the score obtained by quantifying the features of each layer and the average value of the scores across all multiple layers. However, the above-described method of calculating the degree of discrepancy is merely an example, and other methods may also be used to calculate the degree of discrepancy.

[0039] Next, the training data selection unit 202 excludes layers whose deviations are equal to or greater than a predetermined value and selects the other layers (i.e., layers whose deviations are less than the predetermined value) as training data (step S32). The predetermined value here is a threshold value for determining whether the deviations are large enough to be determined as inappropriate data for reflecting the characteristics of an individual user. The predetermined value may be set as an appropriate value obtained in advance, for example, by prior simulation. Note that if there is no layer whose deviation is equal to or greater than the predetermined value (i.e., if the deviations of all layers are less than the predetermined value), the training data selection unit 202 may select all layers as training data.

[0040] Alternatively, the training data selection unit 202 may select training data by accepting a user operation instead of selecting a layer whose deviation is less than a predetermined value as training data. That is, the training data selection unit 202 may allow the user to select a layer to be used as training data from among layers whose deviation is less than a predetermined value, and may select the layer selected by the user operation as training data.

[0041] Thereafter, the learning unit 203 reads out the user data selected by the learning data selection unit 202 from the user data storage unit 201 (step S12). Then, the learning unit 203 executes learning of the PGAI 10 using the read out user data as learning data (step S13). When the learning 203 is completed, the learning unit 203 saves the learned model (step S14).

[0042] (Technical Effects) Next, the technical effects obtained by the learning system 1 according to the third embodiment will be described.

[0043] As described in FIG. 5, in the learning system 1 according to the third embodiment, user data to be used for learning is selected after excluding stratums with a deviation degree equal to or greater than a predetermined value. In this way, it is possible to automatically exclude user data that is not suitable for reflecting the user's characteristics and to learn the PGAI 10. For example, suppose that the user's health condition was poor at some time, and user data that is far removed from the user's characteristics was stored. Such user data is considered to have a high degree of deviation from other user data, and therefore is not selected as learning data in the learning system according to this embodiment. In this embodiment, the use of exceptional user data as described above can prevent inappropriate learning from being performed.

[0044] This disclosure is not limited to the above-described embodiments, but may be modified as appropriate within the scope of the claims and the gist or concept of the invention as can be read from the entire specification, and learning systems involving such modifications are also included within the technical scope of the present invention. [Explanation of symbols]

[0045] 1. Learning System 10 PGAI 20 Processing equipment 201 User data storage unit 202 Learning Data Selection Unit 203 Learning Department

Claims

1. a storage means for storing user data acquired from a user in a plurality of layers; a selection means for selecting a layer to be used as training data for a generative model from the plurality of layers; training means for training the generative model to be personalized to the user using the user data included in the selected layer; A learning system that includes:

2. The storage means stores the user data in layers for each predetermined period. The learning system of claim 1 .

3. The storage means stores the user data in layers according to the user's affiliation. The learning system of claim 1 .

4. the selection means selects a layer to be used as the training data in response to a selection operation by the user. A learning system according to any one of claims 1 to 3.

5. the selection means compares the user data between the plurality of layers, excludes a layer whose deviation from other layers is greater than a predetermined value, and selects a layer to be used as the training data. A learning system according to any one of claims 1 to 3.

Images