Information processing device and program

Abstract

This invention provides an information processing device and program that can present the basis for the decisions made by a learning model. [Solution] The information processing device includes a storage unit having an analogy model that includes an attention structure for weighting multiple pieces of information and a learning model for performing predetermined inferences on the multiple pieces of information weighted by the attention structure, and a display control unit (control unit) that causes the attention structure to display the weighted values ​​assigned to the multiple pieces of information on a display unit, the display control unit causing the display unit to display the weighted values ​​assigned to the multiple pieces of information as a heat map or to display the time change of the weighted values ​​assigned to the multiple pieces of information.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus and a program that perform a predetermined inference from a plurality of pieces of information using a learning model.

Background Art

[0002] Patent Document 1 discloses a technique for performing inference on a plurality of data using a learned learning model.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the technique described in Patent Document 1, it was unclear which information was emphasized and to what extent in the judgment of the learning model. That is, the basis for the judgment of the learning model, such as what changes in the situation would cause the judgment of the learning model to change, was not clear. An object of the present invention is to be able to present the basis for the judgment of the learning model.

Means for Solving the Problems

[0005] To achieve the above object, one embodiment of the present invention is an information processing apparatus, an attention structure that assigns weights to a plurality of pieces of information, a learning model that performs a predetermined inference on the plurality of pieces of information weighted by the attention structure, and a display control unit that causes the display unit to display the weighting values assigned by the attention structure to the plurality of pieces of information. The information processing apparatus includes the above components.

Effects of the Invention

[0006] According to the present invention, weighted values ​​assigned by an attention structure to multiple pieces of information input to the learning model are displayed on the display unit. This makes it possible to show which pieces of information were given more weight in the learning model's decision-making, in other words, to present the basis for the learning model's decision. [Brief explanation of the drawing]

[0007] [Figure 1] This is a block diagram showing the schematic control configuration of the information processing device according to the embodiment. [Figure 2] This is a diagram illustrating the analogy model according to the embodiment. [Figure 3] This is a flowchart showing the information processing flow according to the embodiment. [Figure 4] This figure shows an example of how the attention structure according to the embodiment displays weighted values ​​assigned to multiple pieces of information. [Modes for carrying out the invention]

[0008] Embodiments of the present invention will be described below with reference to the drawings.

[0009] [Configuration of the information processing device] Figure 1 is a block diagram showing a schematic control configuration of the information processing device 1 according to this embodiment. The information processing device 1 according to this embodiment is a computer that performs predetermined inferences from multiple pieces of information using an inference model 30 that includes a learning model 32. Specifically, as shown in Figure 1, the information processing device 1 includes an input unit 22, a display unit 23, a storage unit 26, and a control unit 27.

[0010] The input unit 22 is an input means for the user to input various operations to operate the information processing device 1, and includes, for example, a pointing device such as a mouse or a keyboard. The display unit 23 is composed of, for example, a liquid crystal display, an organic EL display, or other type of display. The display unit 23 displays various information based on display signals from the control unit 27. The display unit 23 may also be a touch panel that also functions as part of the input unit 22, or it may provide audio output.

[0011] The memory unit 26 is a memory composed of RAM (Random Access Memory), ROM (Read Only Memory), etc. The memory unit 26 stores various programs and data, and also functions as a work area for the control unit 27. The memory unit 26 has a program 261 pre-stored in it for executing the information processing described later (see Figure 3).

[0012] Furthermore, the memory unit 26 has pre-stored an analogy model 30 that performs a predetermined inference from multiple pieces of information. The analogy model 30 is pre-built using machine learning with predetermined training data and stored in the memory unit 26. As shown in Figure 2, the analogy model 30 includes an attention structure (attention mechanism) 31 and a learning model 32. The attention structure 31 is a mechanism for weighting multiple pieces of information (data) 35. For example, the attention structure 31 assigns a smaller weight to information 35 that is not important for the inference (decision) of the learning model 32 compared to the information 35 that is important for that inference. The attention structure 31 is generated when the learning model 32 is trained, and the weights of the information that should be emphasized in that situation are learned. The learning model 32 performs predetermined inferences on multiple pieces of information 35 weighted by the attention structure 31. The learning model 32 may be, for example, an AI (Artificial Intelligence) having a deeply trained neural network.

[0013] As shown in FIG. 1, the control unit 27 is constituted by, for example, a CPU (Central Processing Unit) or the like, and controls the operations of each part of the information processing apparatus 1. Specifically, the control unit 27 operates each part of the information processing apparatus 1 based on the operation content of the input unit 22 or the like, develops a program stored in advance in the storage unit 26, and executes various processes in cooperation with the developed program.

[0014] [Information Processing Flow] FIG. 3 is a flowchart showing the flow of information processing by the information processing apparatus 1, and FIG. 4 is a diagram showing an example of display of weighted values assigned by the attention structure 31 to a plurality of pieces of information 35. In the information processing of the present embodiment, a predetermined analogy is made from a plurality of pieces of information by the analogy model 30. This information processing is executed, for example, when a predetermined execution operation is input by the user, by the control unit 27 reading out and developing the program 261 from the storage unit 26.

[0015] As shown in FIG. 3, when the information processing is executed, first, the control unit 27 acquires a plurality of pieces of information 35 that are input data (step S1). The plurality of pieces of information 35 may be those stored in the storage unit 26 in advance, or may be acquired from the outside through a communication unit or the like. Also, the data type and the like of the plurality of pieces of information 35 are not particularly limited.

[0016] Next, the control unit 27 performs weighting on the plurality of pieces of information 35 acquired in step S1 by the attention structure 31 of the analogy model 30 (step S2). The attention structure 31 assigns a weighted value (weight) to each piece of information 35 based on the importance affecting the judgment of the learning model 32.

[0017] Next, the control unit 27 causes the display unit 23 to display the weighted values assigned by the attention structure 31 to the plurality of pieces of information 35 in step S2 (step S3). Here, the control unit 27 extracts information on the weighting values for each piece of information 35 from the attention structure 31 and causes the display unit 23 to display it. The display mode in this case is not particularly limited. For example, as shown in FIG. 4, it may be a heat map that visualizes the magnitude of the weighting value by hue. Also, here, it is preferable to display the temporal change (time series) of the weighting value. In this way, in step S3, it is presented to the user which piece of information 35 was emphasized in the determination of the learning model 32, that is, the basis for the determination of the learning model 32.

[0018] Next, the control unit 27 performs a predetermined inference by the learning model 32 on the plurality of pieces of information 35 weighted by the attention structure 31 in step S2 (step S4). Next, the control unit 27 causes the display unit 23 to display the inference result by the learning model 32 (step S5). Then, the control unit 27 ends the information processing.

[0019] [Technical effects of this embodiment] As described above, according to this embodiment, the weighting values given by the attention structure 31 to the plurality of pieces of information 35 input to the learning model 32 are displayed on the display unit 23. Thereby, it is possible to present to the user which information was emphasized in the determination of the learning model 32, that is, the basis for the determination of the learning model 32. Also, by the user confirming the weighting learned by the learning model 32, it is possible to distinguish between the information necessary and the unnecessary information when the learning model 32 makes a determination. Also, by comparing in various situations which information the learning model 32 emphasizes in making a determination, it is possible to determine which information the learning model 32 is recognizing the situation based on. Also, during the operation of the learning model 32, the user can confirm the basis for the determination of the learning model 32, check whether the learning model 32 is looking at the information that must be emphasized, and determine whether the determination result of the learning model 32 can be used. Furthermore, by comparing the weighting of information from multiple learning models 32, it is possible to compare the information 35 that each learning model 32 emphasizes, and to identify the characteristics of each learning model 32.

[0020] Furthermore, according to this embodiment, the weighted values ​​assigned to the multiple pieces of information 35 are displayed in a heatmap. This allows users to immediately recognize, simply by looking at the heatmap, which information was given more weight and to what extent in the learning model 32's decision-making process.

[0021] Furthermore, according to this embodiment, the time change (time series) of the weighted values ​​assigned to the multiple pieces of information 35 is displayed. This makes it possible to visualize the changes in the basis for the learning model 32's decisions, allowing users to determine when the learning model 32's decisions changed.

[0022] [others] Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above. For example, the application fields and usage patterns of the information processing in the above embodiment are not particularly limited. For example, it can be used in design methods that handle multiple design parameters, or as a human-machine interface for autonomous driving. In the latter case, for instance, the rationale behind the AI's (learning model's) selection of the driving route can be presented to the user, improving the reliability of the AI's decision-making.

[0023] Furthermore, details shown in the above embodiments can be modified as appropriate without departing from the spirit of the invention. [Explanation of Symbols]

[0024] 1. Information Processing Device 23 Display section 27 Control Unit (Display Control Unit) 30 Analogical Models 31 Attention Structure 32 Learning Models 35 Information 261 Programs

Claims

1. An attention structure that weights multiple pieces of information, A learning model that performs predetermined inferences on the plurality of pieces of information weighted by the attention structure, The attention structure includes a display control unit that causes the display unit to display the weighted values ​​assigned to the plurality of pieces of information, An information processing device equipped with the following features.

2. The display control unit displays the weighted values ​​assigned to the plurality of pieces of information in a heat map. The information processing apparatus according to claim 1.

3. The display control unit displays the time change of the weighted value assigned to the plurality of pieces of information. The information processing apparatus according to claim 1.

4. Computers, An attention structure that weights multiple pieces of information. A learning model that performs predetermined inferences on the plurality of pieces of information weighted by the attention structure, A display control unit that causes the attention structure to display the weighted values ​​assigned to the plurality of pieces of information on the display unit. A program that makes it function as such.

Images