JP2019028993A - Information processing device, information processing method, and program - Google Patents

Abstract

To easily change knowledge and feelings.SOLUTION: An information processing device includes input means (interface) for inputting information, a plurality of knowledge cells having input definition information and output definition information to output the output definition information in the case that information having a high correlation to the input definition information is inputted, and output means (interface) for outputting the information outputted from the knowledge cells. A knowledge cell inputs information outputted from the other knowledge cells together with the information inputted from the input means.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an information processing apparatus, an information processing method, and a program.

  A neural network is a model conceived to simulate brain neurons and synapses, and is configured by two-stage processing of learning and identification. In the learning stage, features are learned from a large number of inputs, and a neural network for identification processing is constructed. In the identification stage, the neural network is used to identify what the new input is.

  In recent years, the technology at the learning stage has been greatly developed. For example, a deep neural network having high expressive power can be constructed by deep learning (deep learning).

  The knowledge structure of a neural network can be said to be an optimal knowledge structure because the processing speed is high when inference is performed as knowledge in which the structure of conceptual knowledge does not change. Moreover, the knowledge of the feature concept becomes deeper as more parent data is learned, and it can be said that it is an excellent knowledge reasoning system when evaluating the entire concept knowledge.

JP 2010-146514 A

  However, in a neural network, it is difficult to map a knowledge model in which the definition and structure of conceptual knowledge changes with time and events. For example, since it takes a lot of time to change the concept knowledge that is connected in the network, there is a problem that it is difficult to change the thinking mechanism by correcting the stored knowledge as in the case of humans. .

  In the case of human beings, emotions change from moment to moment depending on the surrounding environment and relationships with people, but neural networks have the problem that it is difficult to build emotion models that change in this way.

  An object of the present invention is to provide an information processing apparatus, an information processing method, and a program capable of easily changing knowledge.

In order to solve the above problems, the present invention has an input means for inputting information, input definition information, and output definition information, and when information having high correlation with the input definition information is input. A plurality of knowledge cells that output the output definition information; and an output unit that outputs information output from the knowledge cell to the outside. The knowledge cell includes other information input from the input unit and other information. Information output from the knowledge cell is also input.
According to such a configuration, an information processing apparatus capable of easily changing knowledge can be provided.

Further, the present invention is characterized in that the knowledge cell has a plurality of the input definition information, and outputs the output definition information when at least one of the plurality of definition information is input.
According to such a configuration, highly relevant knowledge can be output.

Further, the present invention is characterized in that the knowledge cell can set a high degree of correlation between information input from the input means or another knowledge cell and the input definition information.
According to such a configuration, it is possible to arbitrarily set the degree of relevance by setting the degree of correlation.

In the present invention, the knowledge cell includes a plurality of the input definition information, and when a plurality of the input definition information is input, a condition for outputting the output definition information can be set. Possible.
According to such a configuration, a flexible system can be configured by setting conditions for outputting the output definition information.

In addition, the present invention is characterized by having a modification means for adding, deleting, or correcting the knowledge cell.
According to such a configuration, knowledge can be freely added, deleted, and corrected.

Further, the present invention is characterized in that the input definition information and the output definition information are composed of text information, audio information, or image information.
According to such a configuration, various information can be accumulated as knowledge.

In addition, the present invention includes an input step for inputting information, input definition information, and output definition information. When information having a high correlation with the input definition information is input, the output definition information is output. A plurality of knowledge cells, and an output step of outputting the information output from the knowledge cell to the outside, wherein the knowledge cell includes information input from the input step and information output from the other knowledge cells Is also input.
According to such a method, an information processing apparatus capable of easily changing knowledge can be provided.

The present invention also includes a computer having an input step for inputting information, input definition information, and output definition information. When information having a high correlation with the input definition information is input, the output definition A plurality of knowledge cells for outputting information, and an output step for outputting the information output from the knowledge cell to the outside, wherein the knowledge cell is output from the other knowledge cells together with the information input from the input step. According to such a program characterized in that information to be input is also input, an information processing apparatus capable of easily changing knowledge can be provided.

  ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to provide the information processing apparatus, the information processing method, and program which can change knowledge easily.

It is a figure which shows the activation site | part when hearing a chord, and a brain mechanism. It is a figure for demonstrating operation | movement of the neuron knowledge model which concerns on 1st Embodiment of this invention. It is the figure which compared deep learning and knowledge reasoning system. It is a figure for demonstrating the specific operation | movement of 1st Embodiment. It is a figure for demonstrating the specific operation | movement of 1st Embodiment. It is a figure for demonstrating the specific operation | movement of 1st Embodiment. It is a figure for demonstrating the specific operation | movement of 1st Embodiment. It is a figure for demonstrating operation | movement of the emotion reasoning system which concerns on 2nd Embodiment of this invention. It is a figure for demonstrating the specific operation | movement of 2nd Embodiment. It is a figure for demonstrating the specific operation | movement of 2nd Embodiment. It is a figure for demonstrating the succession concierge of the special technique which concerns on the deformation | transformation embodiment of this invention. It is a figure for demonstrating the visitor's emotion learning communicator which concerns on the deformation | transformation embodiment of this invention. It is a figure for demonstrating the caregiver's mind care which concerns on the deformation | transformation embodiment of this invention. It is a figure which shows the structural example of the information processing apparatus which concerns on embodiment of this invention. It is a figure for demonstrating the detail of the knowledge cell which the information processing apparatus which concerns on embodiment of this invention has. It is a figure which shows the specific structural example of the knowledge cell shown in FIG. It is a figure which shows the relationship between the knowledge cell shown in FIG. 15, and input information. It is a figure which shows a mode that the intensity | strength shown in FIG. 15 is set. It is a figure which shows the mode of the cooperation awakening of a knowledge cell. It is a figure which shows the mode of the concrete cooperation awakening of a knowledge cell. It is a figure which shows the independence of the knowledge cell of this invention. It is a figure which shows an example of the system which does not have independence. It is a figure which shows a mode that a knowledge cell is added. It is a figure which shows a mode that a knowledge cell is deleted. It is a figure which shows a mode that a knowledge cell is corrected. It is a figure which shows schematic structure of the information processing apparatus which concerns on embodiment of this invention. FIG. 27 is a diagram for describing an operation of the information processing apparatus illustrated in FIG. 26. FIG. 27 is a diagram for describing an operation of the information processing apparatus illustrated in FIG. 26.

  Next, an embodiment of the present invention will be described.

First embodiment

  1. Proposal of neuron knowledge model based on brain-type knowledge structure

1. Introduction The neuron knowledge model is considered to have brain-type knowledge structure as the starting point. It is a synapse of a mechanism that transmits information and a neuron that has a function to store information. A neural network is basically based on a similar memory model, but there are two major differences. It is a method of reasoning for freedom of memory and knowledge. By studying these two, we were able to propose the mechanism of human memory as knowledge inference of brain-type knowledge structure, and realized an emotion inference system based on a neuron knowledge model. However, not all the mechanisms of human memory have been proved by the neuron knowledge model, but it is in the basic research stage. The results of this research are described below.

  2. Freedom of memory

2.1 Weaknesses of neural networks In neural networks, feature concepts are composed of networks from a superordinate concept to a subordinate concept using a deep tree structure. If this feature concept is considered as one concept knowledge, it can be considered as a knowledge structure in which a plurality of concept knowledges are composed of a deep tree structure.

  The knowledge structure of a neural network can be said to be an optimal knowledge structure with a high processing speed when inference is performed as knowledge in which the structure of conceptual knowledge does not change. The more knowledge of the parent data is learned, the deeper the knowledge of the feature concept becomes, and it can be said that it is a very good knowledge reasoning system when evaluating the whole concept knowledge.

  However, mapping knowledge models in which the definition and structure of conceptual knowledge changes with time and events creates difficulties. For example, it is not impossible to change the concept knowledge connected by the network, but it takes a lot of time, so it is not possible to realize a mechanism in which a person immediately thinks after correcting memory knowledge.

  2.2 Research on knowledge structure freedom

  The degree of freedom of knowledge structure can be derived from consideration of several research themes. Discussions were made on the functional mapping of neuronal neurons and synapses as a knowledge model mechanism: (1) neuronal activation, (2) synaptic signal transmission function, and (3) synapse formation function. .

(1) Activation of neurons It has been explained that neurons activated by neurons are different in the site of activation. For example, FIG. 1 shows an activation site and a brain mechanism when a chord is heard. As shown in the MRI image of FIG. 1, it is described that in music appreciation, a part that feels rhythm is activated in addition to the auditory area. What we want to mention is that the activated sites are separated. If neighboring neurons connected at the synapse are activated, it is possible that continuous neurons activated from the auditory cortex may be displayed on the MRI image, but the activated neurons are separated. It looks like only the part that was. Based on this, the following considerations are made.

<Discussion 1>
As if the related neurons in the brain are activated at a distant place regardless of synaptic signal transmission, the knowledge model also needs a mechanism that allows knowledge to be evaluated regardless of network structure.

(2) Synaptic signal transmission function The synapse of a nerve cell has a function of transmitting a nerve signal by connecting adjacent neurons. In the neural network model, it is considered that the inference processing is continuously transmitted from the upper layer concept knowledge (neuron) to the lower layer concept through the network (synapse). However, the MRI image in FIG. 1 does not appear to synapse continuously transmitting signals. Based on this, the following considerations are made.

<Discussion 2>
Even if neurons between distantly activated parts in the brain are connected by synapses but not activated, the knowledge model network also conveys information to the concept knowledge without depending on the network A mechanism is necessary.

(3) Synapse formation function In the synapse of nerve cells, a synapse network is actively formed from childhood and various knowledge is generated. It is also known that when a disorder occurs in the brain, a new synaptic network is formed at another site to compensate for the function that has been impaired by the disorder. In other words, the synapse seems to control not only the signal transmission function of neurons but also the shape of memory. Based on this, the following considerations are made.

<Discussion 3>
As if the synapse freely forms a network to the target neuron in the brain by the biological environment, the relationship between the knowledge and concept of the knowledge model is not associative and a mechanism that can always be freely formed is necessary. is there.

  Based on the above three considerations, the conclusion was drawn that a knowledge model with a higher degree of freedom than a neural network model is necessary to realize a brain-type knowledge model.

2.3 Independence of memory The various memory models that have been presented so far have some kind of connectivity with other memories that are strong and weak, and if one memory is unbound, the memory associated with it is extracted. It has been expressed in. However, the conclusion derived from the discussion is not a network-type knowledge model in which the knowledge structure is subject to binding constraints, but forms a semantic relationship while conceptual knowledge exists independently and exchanges information with each other. To realize a knowledge model with a high degree of freedom.

  It is not a new concept that the memory of knowledge and rules is not connected by the conceptual structure determined from the beginning like a neural network, but “the memory itself exists independently without being bound to the network”. The concept of knowledge structure, and this environment is realized by the neuron knowledge model.

  3. Neuron knowledge model

3.1 Simplification of knowledge model In the brain-type knowledge model that has been considered so far, we have attempted to analyze the brain function in detail and construct a knowledge model that covers that function. However, due to the complexity of the model that accompanies the implementation of brain functions that have been newly elucidated and the enlargement of the model processing of brain-type functions, many problems have arisen in the implementation of brain-type knowledge models. These causes are largely due to the fact that the brain-type knowledge model has been designed from the viewpoint of brain function. The neuron knowledge model focuses on designing a knowledge model with a simple mechanism by modeling neurons and synapses of neurons that are the basis of brain function. As a result, the memory function can be realized with a simple and highly flexible knowledge structure model.

3.2 Memory function In the brain, memory is formed by neurons and synapses. The memory is operated with various information and knowledge with the three major functions of “memorization”, “save” and “recall”. The memory function is also established in the neuron knowledge model.

(1) Memory recall and memorization In the neuron knowledge model, memory (knowledge) recall is initiated by external stimuli such as vision, hearing and sensation. If there is no memory to be recalled, the external stimulus information is recorded as memory.

(2) Inflection of memory The inference results differ depending on the temporal and environmental factors of the memory held in the neuron knowledge model because the inflection of activation changes due to changes in the external environment.

(3) Memory preservation The knowledge density of memory according to the neuron knowledge model always changes, but different memories may have the same meaning in the initial stage. When the memory is saved in this state, the outside is requested to learn the difference knowledge.

3.3 Knowledge Cell (Mechanism of Knowledge Model)
FIG. 2 is a diagram illustrating a configuration example of the neuron knowledge model according to the first embodiment of the present invention. The unit of knowledge in the neuron knowledge model is called a knowledge cell and functions as a neuron of a neuron. The memory mechanism of the neuron knowledge model can be explained by the function of the knowledge cell, as the brain memory is formed by neurons.

(1) Structure of knowledge cell A knowledge cell has an element called “receptor” that responds to a stimulus and an element called “impulse” that transmits the stimulus to another knowledge cell.

(2) Knowledge Cell Memory Knowledge cells also store information from external stimuli, just as neurons are central to brain memory. The knowledge cell has a strong unique sensitivity of knowledge, and when there are other knowledge cells having the same knowledge content, a request for the difference knowledge is transmitted to the outside as a stimulus.

(3) Awakening of knowledge cell (Figure 2- (1))
Just as neurons are activated by signals from synapses, knowledge cells awaken by stimulating receptors. However, since the reception of external stimuli may be suppressed depending on the receptor strength, the knowledge cell does not always wake up in all cases.

(4) Knowledge cell activity (Figure 2- (3))
When a knowledge cell wakes up so that a neuron sends a signal to another neuron through a synapse, it sends a wake-up stimulus from the impulse to another knowledge cell or the outside (response). However, since the transmission of the arousal stimulus may be suppressed depending on the impulse intensity, it is not always transmitted in all cases.

(5) Knowledge cell cooperation awakening (Figure 2- (2))
The arousal stimulus transmitted from the impulse by the awakening of the knowledge cell is transmitted to other knowledge cells. This is called knowledge cell cooperative awakening. Knowledge cell is activated and knowledge is dug up while the receptor repeats the reaction of arousal stimulus by cooperative awakening.

  In the example of FIG. 2, the neuron knowledge model has knowledge V, W, and X as knowledge cells. Knowledge cells have elements called “receptors” that respond to stimuli and elements called “impulses” that transmit stimuli to other knowledge cells. For example, the knowledge cell corresponding to knowledge V has impulse V, receptor A, and receptor B. The knowledge cell corresponding to the knowledge W has an impulse W, a receptor A, and a receptor C. The knowledge cell corresponding to knowledge X has impulse X, receptor V, and receptor B. In FIG. 2, a solid line indicates awakening due to an external stimulus, a broken line with a short interval indicates cooperative awakening due to the awakening stimulus, and a broken line with a long interval indicates a response due to the awakening stimulus.

  In the example of FIG. 2, external stimuli A and B are given to the neuron knowledge model. The receptor A corresponding to the external stimulus A has knowledge cells corresponding to the knowledge V and W. The receptor B corresponding to the external stimulus B has a knowledge cell corresponding to the knowledge V and X. Therefore, when the external stimuli A and B are given, the knowledge cell corresponding to the knowledge V has the receptors A and B, so that the impulse V is activated and the knowledge cell corresponding to the knowledge X having the receptor V is linked. Awaken and returns impulse V. The knowledge cell corresponding to the knowledge W has the receptor A but does not have the receptor B, so the impulse W is not activated and no response is made. In addition, the knowledge cell corresponding to the knowledge X has the receptor B corresponding to the external stimulus B and, as described above, the receptor V is stimulated by the cooperative awakening, so that the impulse X is returned. As a result, in the neuron knowledge model shown in FIG. 2, impulses V and X are returned to the external stimuli A and B, respectively. Note that the number of knowledge cells, the number of impulses and receptors included in each knowledge cell shown in FIG. 2 are examples, and other numbers may be used.

3.4 Goal of cooperative awakening When people evoke memory, there are cases where the memory recall ends to some extent, and where the memory is digged deeply until the desired event is reached. Knowledge cell cooperative awakening in a neuron knowledge model can dig up memory many times, but at what level of memory awakening ends depends on the user of the system environment.

  4). Knowledge reasoning

4.1 Definition of knowledge reasoning Deep learning system that infers with deep tree structure knowledge is called machine reasoning, but in a broad sense it is knowledge inference mechanism because it infers with knowledge. However, in the present specification, a mechanism inferred by a neuron knowledge model that infers with knowledge having a high degree of freedom is defined as knowledge inference as a mechanism opposite to machine inference.

4.2 Comparison between Deep Learning and Knowledge Reasoning FIG. 3 shows a comparison table between deep learning and knowledge inference systems using a neuron knowledge model.

4.3 Positioning in the system environment Since knowledge inference is an inference mechanism based on a neuron knowledge model, it is necessary to have a system environment in which external stimuli are received from the input interface and inference results are sent from the output interface to the outside. For example, when system products are based on voice input, artificial intelligence technology for speech recognition and semantic understanding is required, when using camera image input, artificial intelligence technology for image recognition is required, and for speech output, artificial intelligence technology for voice text generation I / O interface technology is required.

4.4 Embodiment of Knowledge Inference A specific operation of the first embodiment of the present invention will be described below. 4-7 is a figure explaining the specific operation | movement of 1st Embodiment. 4 to 7, {} indicates a knowledge body of the knowledge cell, * indicates an impulse (an element that emits a stimulus by awakening), and: indicates a receptor (an element that passively stimulates a stimulus). The input conditions are as follows.

(1) Input conditions (image)
Input format for knowledge reasoning (tentative)
““ Main part ”is“ [Modification part] ”“ Target part ”
Input instructions to knowledge reasoning At the end! Is the learning input instruction at the end? The question mark is a question input prompt Prompt ● User prompt Prompt ◎ Prompt for knowledge answer Prompt for user Knowledge prompt

  In addition, Japanese analysis and sentence comprehension are realized by separate functions.

  In FIG. 4, the question “white dog” wakes up the “Spitz” knowledge in response to the receptor as an external stimulus. The “Spitz” impulse then sends a stimulus and responds.

  In FIG. 5, when learning “Kishu dog is a white dog” and storing the memory, in order to understand that the knowledge of “white dog” also applies to “Spitz” Ask guidance questions about the difference between Kishu Dog and Spitz.

  In FIG. 6, the question “white dog” reacts to the receptor as an external stimulus, and the “Spitz” and “Kishu dog” knowledge awakens. However, since the receptor “long hair” has not yet reacted to “Spitz”, a narrow-down question is asked.

  In FIG. 7, there is no “white” receptor in the “animal” knowledge. When the “animal” in the target section is evaluated as an external stimulus, the “dog” and “bird” knowledge awakens. However, since there is no “white”, “bird” and “dog” are awakened as awakening stimuli. Then we understand that “white” receptors exist in “Swan”, “Spitz” and “Kishu Dog”. From the first question, the question is narrowed down to the “flying” receptor that is not responding to differentiate between “birds” and “dogs”.

Second embodiment

  Emotion reasoning system based on neuron knowledge model

  Human emotions are constantly changing depending on the surrounding environment and relationships with people. If emotions are handled by artificial intelligence, an emotional knowledge model that can follow emotional changes is necessary. By the way, when people feel things, they express their emotions with a mechanism in their brain. Neurons that are neurons in the brain and synapses work to express complex emotions. In order to push the ability of artificial intelligence to the area that handles human feelings and emotions, a mechanism that can handle complex emotion expressions with simple models is required.

  Therefore, using the knowledge base of the neuron knowledge model proposed in the previous chapter, we studied a model for inferring emotional knowledge, and realized a system for inferring emotions. However, not all the emotional mechanisms of human beings have been verified by the emotion reasoning system, but they are in the basic research stage.

6.1 Emotional expression model A person is moved by an external stimulus. The biological mechanism from the stimulation to the movement of the heart involves complex communication of signals within the brain. Previous attempts to handle emotions with artificial intelligence have aimed to model the complex and biological mechanisms that form these emotions. However, when objectively looking at the emotions in the final state where the mind has moved, emotional expression means that when an external stimulus is applied, the memorized emotions are activated through some biological action in the brain. It can be said that it has changed.

6.2 Mechanism for inferring emotions The actions of the human mind are very complex. Positive and negative emotions can be mixed and mixed. However, this study is not a study of biological models that express complex biological actions of the mind such as joy and sadness. This is a study of a mechanism that evokes emotions as a memory in response to emotional stimuli from the outside and infers against collective common emotions. This is because it is possible to infer emotional knowledge from sensory information given as external stimuli by considering the memorized emotions as knowledge.

6.3 Emotional Inference by Neuron Knowledge Model The neuron knowledge model in the previous chapter infers while activating the learning knowledge. If an environmental model that activates the knowledge of this model as memorized emotional knowledge can be inferred, emotional knowledge can be inferred. The method of inferring emotional knowledge by this mechanism is defined as emotional reasoning.

  7). Emotion inference

7.1 Definition of Emotion Inference Emotion inference is a method of inferring the emotional knowledge being learned from emotions and sensory information acquired as external stimuli using the knowledge base of the neuron knowledge model.

(1) Purpose of emotional reasoning Emotional reasoning can provide a communication environment that understands emotions by searching for emotional knowledge that you have memorized and sympathizing with the emotional knowledge of others, and can care for the transmission of your mind. The purpose is to contribute to society with a human-friendly system.

(2) Features of Emotion Reasoning Emotion reasoning is broadly divided into “self-emotion reasoning” and “altruistic emotion reasoning” depending on the type of knowledge handled in reasoning. Self-emotion reasoning uses emotion reasoning in one's life, such as self-development and self-reflection, by making inferences from the emotional knowledge memorized in one's own experience. In the case of altruistic emotion inference, emotions are inferred from other people's experiences and emotional knowledge stored in collective learning. This uses emotion reasoning as a specific environment or tool, such as assisting emotional communication with others, analyzing popular emotions, or searching for emotions using collective emotional knowledge.

(3) Formation of sensibility When a person feels an emotion by an external stimulus, the state of sensibility at that time greatly affects. In emotional reasoning, when inferring emotional knowledge, it is not directly inferred by emotions received from external stimuli, but an emotional reasoning environment is prepared in which the sensation at the time of inference is first formed. As a result, emotional knowledge can be inferred from the state of the mind when performing emotional reasoning. Sensitivity is made up of pieces of emotional memory called “pleats”.

(4) Sensitivity knowledge Sensitivity formed from external stimuli is constructed as sensitivity knowledge. This sensibility becomes a design drawing of one's own mind, and whenever a sensibility is formed from an external stimulus, a stored sensibility close to the stimulus is called and used. Sensitivity knowledge is basically constructed for each user, but may refer to the sensitivity of others as needed.

(5) Inference and Learning of Emotional Knowledge Emotional knowledge is inferred in the same way as a knowledge inference model by the sensibility formed as an emotional inference environment. When there is no emotional knowledge activated by sensitivity, sensitivity information is learned as emotional knowledge.

8). Emotion Reasoning System 8.1 Knowledge Model of Emotion Reasoning The emotion reasoning system is an inference system built using emotion reasoning. The reasoning mechanism will be described below.

  FIG. 8 is a diagram showing the movement of knowledge when emotion inference is performed. Since the emotion reasoning system is also based on a neuron knowledge model, the reasoning behavior of knowledge is the same.

(1) Structure of knowledge The emotion reasoning system differs from the knowledge reasoning model in that there are two reasoning mechanisms. One is a "Kansei knowledge" reasoning mechanism that makes inferences when forming one's own sensibility, and the other is an "Emotion knowledge" reasoning mechanism that makes inferences about target emotional knowledge. In the case of “altrusive emotional reasoning” that infers emotions of others and emotional knowledge built by collective learning, it is necessary to separate emotional knowledge from emotional knowledge, but in the case of constructing “selfish emotional reasoning” both Can be treated as common emotional knowledge.

(2) Emotional knowledge cell (structure of knowledge)
Since the knowledge environment of the emotion reasoning system is based on a neuron knowledge model, the unit of knowledge handled by emotion reasoning is also a knowledge cell. The emotion reasoning system is composed of two types of knowledge. The “pleats” of sensitivity knowledge and the “knowledge” of emotion knowledge are emotion-type knowledge cells. An emotional knowledge cell has an element called “sense” that receives emotional stimuli from the outside and an element called “sympathy” that transmits emotional stimuli. The sense element has a mechanism that suppresses reception of a stimulus called sense intensity, and the emotional knowledge cell is not always activated in all cases. In addition, the sympathy element may suppress the transmission of emotional stimuli depending on the sympathy strength.

(3) Formation of Kansei When the emotion inference system is externally stimulated (Fig. 8- (1)), it forms its own sensibility (mind pleats) that is the state of mind for emotion inference (Fig. 8- (2)). When emotions arouse, a person affects the state of the mind, recalls the memorized situation, and forms a trap of mind. The emotion reasoning system infers sensibility knowledge that memorizes past external stimuli, acquires pleats (mind's heels), and forms sensibilities with the sympathy of pleat information. The inference of sensibility knowledge at this time follows the inference method of the neuron knowledge model.

(4) Activation of Knowledge Inference from Kansei The formed Kansei activates the inference function of emotional knowledge as an external stimulus (FIG. 8- (3)). In response to an external stimulus from the sympathy that formed self-sensitivity (heartbeat), the sympathy of the emotional knowledge activated by sense is the external interface (Reply: Fig. 8- (4)) and other emotional knowledge Send awakening stimulus to the cell. This activates the knowledge stored in emotional knowledge and responds. The inference of emotional knowledge at this time follows the inference method of the neuron knowledge model.

8.2 Self-Emotion Reasoning When the emotion reasoning system is used for “self-emotion reasoning”, sensitivity knowledge and emotion knowledge can be handled as common knowledge. This is because the inference range of knowledge is closed by its own emotional memory, so the pleated memory of emotional knowledge that forms sensitivity and the knowledge memory of emotional knowledge that infers emotions are considered to be the same memory of one person. .

8.3 Emotional expression of emotion inference system Emotional knowledge cells deal with various sensibilities. For example, as the sensory sensibility felt by the five senses, sensibility values representing sensations such as “hot”, “cold”, “dazzling”, “suffering”, “quiet” are handled. The emotional sensibility felt by the heart is a sensitivity value representing emotions such as “lonely”, “joyful”, “fun”, and “spicy” (synonym conversion is performed automatically). As described above, when the emotion is constructed by the emotion reasoning system, the human emotional expression is learned as the emotion value of the knowledge cell, instead of dealing with the biological structural expression of the emotion. For example, if you want to infer "fun" and "joyful" as close emotions, you can collect them as mind pleats by learning them as knowledge with the same sense, and gather them in the heart of emotion inference.

  In addition, the example of FIG. 8 has a neuron knowledge model related to sensibility knowledge and a neuron knowledge model related to emotion knowledge. Neuronal knowledge related to sensibility knowledge has an element called “sense” that responds to a stimulus and an element called “sympathy” that forms sensibility by the stimulus. In the example of FIG. 8, the neuron knowledge model relating to sensibility knowledge includes pleats A to C, and pleat A includes sympathy A and senses X and Y.

  In the example of FIG. 8, the neuron knowledge model related to emotional knowledge has an element called “sense” that reacts to a stimulus and an element called “sympathy” that forms sensitivity by the stimulus. In the example of FIG. 8, the neuron knowledge model related to emotional knowledge has knowledge V to X, and knowledge V has sympathy V and senses A and F.

  As shown in FIG. 8, when external stimuli X and Y are given, a pleat having a sense corresponding to the external stimuli X and Y of the neuron knowledge model related to the sensibility knowledge reacts and a state of mind for performing emotion inference Mind pleats are generated as the sensibility of the self. In the example of FIG. 8, mind pleats having sympathies A to F are generated.

  When mind pleats are generated, a knowledge cell having a sense corresponding to sympathy A, C, F reacts and responds in a neuron knowledge model related to emotional knowledge. In the example of FIG. 8, sympathies V and X are returned. Note that the number of pleats related to sensitivity knowledge and the number of knowledge related to emotion knowledge shown in FIG. 8 are examples, and other numbers may be used.

4.4 Embodiment of Emotion Inference A specific operation of the second embodiment of the present invention will be described below. 9 and 10 are diagrams for explaining a specific operation of the second embodiment. First, FIG. 9 shows an operation example of emotion inference (kimono selection) based on kimono knowledge. In the example of FIG. 9, “adult love playing in the nature of the river” is given as an external stimulus, and this external stimulus works on the sensibility knowledge as an individual's sensibility memory, forming a sensibility as a mind pleat. The mind pleat formed in this way works on emotional knowledge as kimono knowledge, and outputs the type of kimono as an inference result. That is, in the example of FIG. 9, emotional sensibility is formed from an external stimulus, and the sense of emotional knowledge of kimono is inferred from emotion.

  FIG. 10 shows an operation example of emotion inference (event planning) based on event knowledge. In FIG. 10, “playing at an amusement park” is given as an external stimulus, and this external stimulus works on the sensibility knowledge as an individual's past sensibility memory, forming a sensibility as a mind pleat. The mind pleat formed in this way works on emotional knowledge as event knowledge, and outputs the type of event as an inference result. That is, in the example of FIG. 10, in the formed sensibility, the sympathy performs control not to receive the cooperative stimulus by suppressing the sense intensity of emotional knowledge.

  Next, an implementation example of the emotion inference system according to the second embodiment will be described.

9. Research Implementation of Emotion Reasoning System As of July 2017, we will introduce a system where research and implementation of knowledge reasoning and emotion reasoning are in progress.

(1) Succession of specialized technology Concierge (manufacturing company)
-Succession of specialized technology through knowledge dialogue of knowledge reasoning-
FIG. 11 is a diagram for explaining the inheritance concierge of specialized technology. In FIG. 11, the experience and skill of the three specialized techniques are converted into knowledge so that anyone can use them. Each knowledge is constantly updated as laws change.

(2) Emotion learning communicator (Aquarium)
-Scale survey by participatory emotion acquisition in emotion reasoning-
FIG. 12 is a diagram for explaining a visitor's emotion learning communicator. In FIG. 12, the emotional scale is learned by planning dialogue with the visitors, and emotional knowledge is aggregated and utilized for real-time display improvement and planning.

(3) Caregiver's mind care environment (care facility)
〜Mental transmission by emotion filter of emotion reasoning〜
FIG. 13 is a diagram illustrating mind care for the care recipient. In FIG. 13, the difference between the sensitivity formed from the emotion expressed by the care recipient and the sensitivity understood by the caregiver is cared for, and an environment in which the mind can be transmitted is provided. Sensitivity from caregivers is also taken care of and communicated.

Modified embodiment

  A modified embodiment will be described below.

(1) Changes in knowledge density with time ・ Research on changes in knowledge element strength ・ Research on survival cycle of knowledge cells

(2) Verification of emotional structure and cooperative awakening ・ Research on emotional expression in complex systems ・ Study on modeling of emotional transition

(3) Cooperation of external stimulus interface ・ Verification research on input / output interface ・ Development research on interface infrastructure

(4) Cooperation with neural network knowledge (hybrid knowledge model)
・ Research on knowledge linkage of knowledge structure ・ Research on engine linkage of inference mechanism

  The inventor of the present application has no reason to criticize the existing reasoning mechanism, and conversely thinks that a system in which the existing reasoning mechanism is optimal needs to have a more advanced goal and be implemented. Among them, the neuron knowledge model is one of the tools to prove the possibility of artificial intelligence while complementing the existing reasoning mechanism.

  Also, not all of the brain and emotional mechanisms have been demonstrated, and research work continues on issues. If possible, we hope that it will become the foundation for the future development of the artificial intelligence research field by pursuing the potential of neuron knowledge models in cooperation with research institutions with a wide field of view.

  And we are going to advance the neuron knowledge model to the next inference world of the emotion inference system.

  9.1 Description of Configuration of Embodiment of the Present Invention

  Next, an embodiment of the present invention will be described. FIG. 14 is a diagram illustrating a configuration example of the information processing apparatus according to the embodiment of the present invention. As shown in FIG. 14, an information processing apparatus 10 according to an embodiment of the present invention includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, and a HDD (Hard Disk Drive). 14, a GPU (Graphical Processor Unit) 15, an I / F (Interface) 16, a bus 17, and a display device 18.

  Here, the CPU 11 can execute processing based on programs stored in the ROM 12 and the HDD 14, and can execute processing related to a knowledge cell, which will be described later, for example.

  The ROM 12 is a semiconductor storage device that stores basic programs executed by the CPU 11 and data.

  The RAM 13 is a semiconductor memory device for temporarily storing a program executed by the CPU 11 and data being calculated.

  The HDD 14 is a magnetic storage device that stores programs executed by the CPU 11 and data.

  The GPU 15 executes a drawing process in accordance with a drawing command supplied from the CPU 11 and supplies the obtained image and video to the display device 18 for display.

  For example, the I / F 16 is an interface that inputs information from an input device such as a keyboard (not shown) and exchanges information with a LAN (Local Area Network) (not shown).

  The bus 17 is a signal line group for connecting the CPU 11, ROM 12, RAM 13, HDD 14, GPU 15, and I / F 16 to each other and enabling exchange of information among them.

  The display device 18 is configured by, for example, an LCD (Liquid Crystal Display) or the like, and displays images, videos, and the like supplied from the GPU 15 on the display unit.

  9.2 Description of operation of embodiment of the present invention

  Next, the operation of the information processing apparatus 10 according to the embodiment of the present invention will be described. The process relating to the knowledge cell described below is realized by the CPU 11 reading and executing a program stored in the HDD 14 shown in FIG.

  FIG. 15 is a diagram illustrating a structure of a knowledge cell handled by the information processing apparatus 10. As shown in FIG. 15, a knowledge cell expressing knowledge has input definition information and output definition information registered in advance. The input definition information is information whose correlation is compared with input information input from the outside. The output definition information is information that defines information to be output when the knowledge cell is activated. These definition information are given different names, the output definition information is “impulse”, and the input definition information is “receptor”.

  FIG. 16 is a diagram illustrating a description example of a knowledge cell. In the example shown on the left side of FIG. 16, the definition information constituting the knowledge cell has “white” and “dog” as input definition information surrounded by a thin line as values, and output definition information surrounded by a thick line. "Kishu Dog" as the value. When such a knowledge cell is described as text information, as shown on the right side of FIG. 16, the output definition information is indicated by a symbol “*”, and the input definition information is indicated by a symbol “:”. And the whole is enclosed by the symbol “{}”. In addition, the said description is an example and it cannot be overemphasized that this invention is not limited to such a case.

  FIG. 17 is a diagram illustrating the operation of the knowledge cell. As shown in FIG. 17, the knowledge cell performs the following information processing when input information is given.

(1) Awakening of knowledge cell When input information is given to the knowledge cell, the correlation between the input information and the input definition information is determined. If it is determined that the correlation is high, the knowledge cell is awakened, When awakened, the process proceeds to the next activation process (described later). Hereinafter, a state where the correlation between the input information and the input definition information is determined to be high is referred to as “a state where the knowledge cell is awakened”.
(2) Activation of knowledge cell When the knowledge cell is awakened by the input information, output definition information is output. This output process is hereinafter referred to as “knowledge cell activation process”.

  More specifically, in FIG. 17, input information A and input information B are input, and the correlation between the input definition information A and the input definition information B is determined. As a result, when it is determined that the correlation between the input information A and the input definition information A is high, and when it is determined that the correlation between the input information B and the input definition information B is high, the knowledge shown in FIG. The cell is awakened.

  When the knowledge cell is awakened, an activation process for outputting information defined by the output definition information is executed. In the example of FIG. 17, output definition information C is output as output information C.

  FIG. 18 is a diagram showing information that can be set in the knowledge cell. As shown in FIG. 18, in the knowledge cell, the following input intensity and output intensity can be set.

(1) Input intensity is information indicating the degree of correlation between input information for waking up a knowledge cell and input definition information, and is set as a real number from 0 to 1. When the intensity is 1, 100% correlation is required, but when awakening with a lower correlation, the value is set lower.

(2) The output intensity is information for setting the output intensity for the output definition information when the knowledge cell is awakened, and is set as a real number from 0 to 1. When the strength is 1, the consistency of all the input definition information is obtained. However, when activation is performed with a small consistency, the numerical value is set lower. That is, the output intensity functions as a threshold for awakening the knowledge cell when there are a plurality of input definition information. As an example, when the output intensity is a small value less than 1, the knowledge cell can be awakened even when only a part of the input information is highly correlated with the input definition information. Conversely, when the output intensity is set to 1, it is possible to wake up only when all input information is highly correlated with the input definition information.

  As a specific example, in the case of (1), in the case of text information, for example, it is possible to awaken a synonym by reducing the intensity. Further, in the case of (2), by setting the strength to less than 1, for example, even when a part of the plurality of input definition information matches the input information, it can be activated.

  FIG. 19 is a diagram for explaining cooperative awakening of knowledge cells. In the knowledge cell according to the present embodiment, output information output by an activation process of a certain knowledge cell can be input information of another knowledge cell. In this case, output information output from a certain knowledge cell is given as input information of another knowledge cell, and when there is corresponding input definition information, the other knowledge cell is awakened. In addition, when other knowledge cells are awakened, output information is output, and yet another knowledge cell having the corresponding input definition information is awakened.

  Thus, in this embodiment, the awakening of the knowledge cell is continued in a chain. Such chained awakening is hereinafter referred to as “knowledge cell cooperative awakening”. In the example of FIG. 19, knowledge cell 1 is awakened by input information A and input information B, and output information C is output. The output information C output from the knowledge cell 1 in this way is given as input information to the knowledge cell 2, and as a result, the knowledge cell 2 is awakened (cooperated). When the knowledge cell 2 is awakened, output information E is output. Such output information E is given as input information to the knowledge cell 3, and the knowledge cell 3 is also awakened (cooperated). When the knowledge cell 3 is awakened, output information F is output. That is, in the example of FIG. 19, the output information C is output by the awakening of the input information A and the input information B, and the output information E and the output information F are output by the cooperative awakening.

  FIG. 20 is a diagram illustrating a specific example of cooperative awakening of knowledge cells. In FIG. 20, the following processing is executed.

(1) When the input information “animal” is input to the knowledge cell 1 and the knowledge cell 1 is awakened, the output information “dog” is output.
(2) The input information “white” is input to the knowledge cell 2, the output information “dog” output from the knowledge cell 1 is input, and the knowledge cell 2 is awakened, so that the output information “Kishu dog” Is output.
(3) “Kishu Dog”, which is output information from the knowledge cell 2, is input to the knowledge cell 3 and awakened, and output information “Wakayama” is output.

  That is, in the example of FIG. 20, “dog” is output by awakening (single) for “white” and “animal” as input information, and “Kishu dog” and “Wakayama” by (cooperation) awakening. Is output. That is, in this embodiment, not only “dog” and “Kishu dog”, which are closely related to the input information “white” and “animal”, but also “Wakayama” are output in cooperation.

  FIG. 21 is a diagram for explaining the specificity based on the independence of the knowledge cell in the embodiment of the present invention. In FIG. 21, since the information cell as a whole does not have a structural connection relationship between knowledge cells, the correction of the knowledge cell does not affect other knowledge cells. For this reason, some knowledge cells can be corrected even when the system is operating.

  On the other hand, FIG. 22 shows a system in which pieces of information are structurally coupled. Such a system has a structural impact on other information when the information is modified. For this reason, when correcting the information, it is necessary to stop the system or reexamine the connection relationship between the information.

  FIG. 23 is a diagram illustrating a case where a knowledge cell is added to the system. As shown in FIG. 23, for example, a knowledge cell 1 and a knowledge cell 2 exist, and a knowledge cell 3 can be added to an operating system by an external addition instruction. At this time, since the knowledge cells do not have a structural connection, the definition information of the knowledge cell 3 by the additional process does not affect other knowledge cells. However, semantic changes occur due to the addition of information.

  FIG. 24 is a diagram for explaining a case where a knowledge cell is deleted. As shown in FIG. 24, in the system in which knowledge cells 1 to 3 exist and are operating, the knowledge cell 3 can be deleted by an external deletion instruction. At this time, since the knowledge cells do not have a structural connection, a structural change that affects other knowledge cells due to the deletion process does not occur. However, semantic changes occur due to information deletion.

  FIG. 25 is a diagram illustrating a case where a knowledge cell is corrected. As shown in FIG. 25, in the system in which knowledge cells 1 to 3 exist and are operating, the knowledge cell 2 can be corrected by an external correction instruction. At this time, since the knowledge cells do not have a structural connection, a structural change that affects other knowledge cells due to the correction process does not occur. However, semantic changes occur due to correction of information.

  FIG. 26 is a diagram illustrating processing for registering a knowledge cell. In the present embodiment, knowledge cells can be added, modified, deleted, and the like by interaction with the user. FIG. 26 shows an example in which a knowledge cell is newly registered. In FIG. 26, the interface has a function of receiving an input from the user and outputting a response from the neuron knowledge model. The interface is configured as, for example, a GUI (Graphical User Interface) displayed on the display device 18 by the GPU 15, and information from the user is input from, for example, a keyboard or a microphone (not shown), and the I / F 16 is input. Can receive. Note that the output from the neuron knowledge model can be displayed on the display device 18 by the GPU 15 or output as sound from a speaker (not shown).

  In the example of FIG. 26, in a state where no knowledge cell is registered, the user inputs “white dog is Kishu dog”. The interface that has received such an input acquires “white dog”, “ha”, and “kishu dog” by, for example, morphological analysis and syntax analysis of “white dog is Kishu dog”. Next, the interface instructs the neuron knowledge model to add knowledge cells. As a result, the neuron knowledge model adds a knowledge cell 1 having “white dog” as input attribute information and “kishu dog” as output attribute information.

  Next, as shown in FIG. 27, after adding knowledge cell 2 having “white bird” as input definition information and “swan” as output definition information, the user sets a “white dog is Spitz”. Suppose that information is entered into the interface. In this case, since there is a knowledge cell 1 having “white dog” as input definition information and “kishu dog” as output definition information, this information and input definition information overlap. Therefore, in order to clarify the difference from the knowledge cell 1, the interface outputs “What is the difference from Kishu Dog?” To the user. As a result, when the user inputs “long hair”, the neuron knowledge model outputs “I understand” to the user via the interface, and “white dog” and “ A knowledge cell 3 having “long hair” and having “Spitz” as output definition information is added to the neuron knowledge model.

  In the state shown in FIG. 27, it is assumed that information on a question “What is a white dog?” Is input from the user to the interface. Such information is supplied to the neuron knowledge model through the above-described morphological analysis processing, syntax analysis processing, and the like. When “white dog” is supplied as input information to the neuron knowledge model, knowledge cell 1 and knowledge cell 3 have “white dog” as input definition information, so knowledge cell 1 and knowledge cell 3 are awakened, Since the knowledge cell 3 has “long hair” as the input definition information, the neuron knowledge model presents “whether the hair is long?” As a question to the user via the interface. As a result, when the user answers “yes”, the knowledge cell 3 having “long hair” as the input definition information is awakened, and “Spitz” is presented to the user as an answer via the interface. .

  10.1 Description of Modified Embodiment

  In the above embodiment, the input definition information and the output definition information of the knowledge cell have only text information. However, as shown in FIG. 28, at least one of the input definition information and the output definition information is You may make it have image information or audio | voice information. In the example of FIG. 28, image information corresponding to the Kishu dog is registered in the knowledge cell, and “Wan Wan” indicating the dog's cry is registered in the knowledge cell. In addition, when comparing the voice information as the input definition information and the voice information as the input information, for example, the voice information is converted into the frequency domain, and the two voice information is compared in the frequency domain. The relationship may be determined. In the case of image information, the object may be detected from the image information by a feature extraction process, and the correlation between the detected objects may be determined.

  Moreover, in the above embodiment, although the case where the number of knowledge cells was 1-3 was mentioned as an example, you may make it have four or more knowledge cells.

  In the above embodiment, the knowledge cell has 1 to 3 pieces of input definition information. However, the knowledge cell may have four or more pieces of input definition information. In the above embodiment, the knowledge cell has one output definition information. However, the knowledge cell may have two or more output definition information.

  Moreover, in the above embodiment, although the input intensity which is the information which shows the correlation degree of the correlation of input information and definition information was set to 1, you may make it designate with the real number between 0-1. Further, when the knowledge cell is awakened, the output intensity as information for setting the intensity for the output information is set to 1, but it may be set as a real number between 0 and 1.

  In the above embodiment, the input intensity is defined as a real number between 0 and 1. However, for example, a negative value (for example, -1) may be included. Thus, a negative correlation can be expressed by including a negative value.

  Further, in the above embodiment, the interface and the neuron knowledge model are configured by the hardware shown in FIG. 14, but other than this, for example, a DSP (Digital Signal Processor) or an FPGA (Field Programmable Gate Array) You may make it comprise by.

10 Information processing apparatus 11 CPU
12 ROM
13 RAM
14 HDD
15 GPU
16 I / F
17 Bus A, B External stimulus V, W, X Knowledge

In order to solve the above problems, the present invention has an input means for inputting information, input definition information, and output definition information, and when information having high correlation with the input definition information is input. A plurality of knowledge cells that output the output definition information; and an output unit that outputs information output from the knowledge cell to the outside. The knowledge cell includes other information input from the input unit and other information. The information output from the knowledge cell is also input, and the knowledge cell can set a high degree of correlation between the information input from the input means or the other knowledge cell and the input definition information. It is characterized by that.
According to such a configuration, an information processing apparatus capable of easily changing knowledge can be provided.

Further, the present invention is characterized in that the knowledge cell has a plurality of the input definition information and outputs the output definition information when at least one of the plurality of the input definition information is input. .
According to such a configuration, highly relevant knowledge can be output.

In addition, the present invention includes an input step for inputting information, input definition information, and output definition information. When information having a high correlation with the input definition information is input, the output definition information is output. A plurality of knowledge cells, and an output step of outputting the information output from the knowledge cell to the outside, wherein the knowledge cell includes information input from the input step and information output from the other knowledge cells And the knowledge cell can set a high degree of correlation between the information input from the input step or another knowledge cell and the input definition information .
According to such a method, an information processing apparatus capable of easily changing knowledge can be provided.

The present invention also includes a computer having an input step for inputting information, input definition information, and output definition information. When information having a high correlation with the input definition information is input, the output definition A plurality of knowledge cells for outputting information, and an output step for outputting the information output from the knowledge cell to the outside, wherein the knowledge cell is output from the other knowledge cells together with the information input from the input step. The knowledge cell is capable of setting a high degree of correlation between the information input from the input step or other knowledge cell and the input definition information. .
According to such a program, an information processing apparatus capable of easily changing knowledge can be provided.

Claims (8)

An input means for inputting information;
A plurality of knowledge cells that output the output definition information, when the input definition information and the output definition information are input and information having high correlation with the input definition information is input;
Output means for outputting information output from the knowledge cell to the outside;
An information processing apparatus, wherein the knowledge cell inputs information output from another knowledge cell together with information input from the input unit. The knowledge cell has a plurality of the input definition information,
Outputting the output definition information when at least one of the plurality of definition information is input;
The information processing apparatus according to claim 1.   The said knowledge cell can set the high correlation of the information input from the said input means or the said other knowledge cell, and the said input definition information, The Claim 1 or 2 characterized by the above-mentioned. Information processing device. The knowledge cell has a plurality of the input definition information,
The information processing apparatus according to claim 2, wherein a condition for outputting the output definition information can be set when a plurality of the input definition information is input.   5. The information processing apparatus according to claim 1, further comprising a modification unit that adds, deletes, or corrects the knowledge cell. 6.   6. The information processing apparatus according to claim 1, wherein the input definition information and the output definition information are configured by text information, audio information, or image information. An input step for entering information,
A plurality of knowledge cells for outputting the output definition information when input definition information and output definition information are input and information highly correlated with the input definition information is input;
An output step of outputting the information output from the knowledge cell to the outside;
An information processing method, wherein the knowledge cell inputs information output from another knowledge cell together with information input from the input step. Computer
An input step for entering information,
A plurality of knowledge cells for outputting the output definition information when input definition information and output definition information are input and information highly correlated with the input definition information is input;
An output step of outputting the information output from the knowledge cell to the outside;
A computer-readable program characterized in that the knowledge cell inputs information output from another knowledge cell together with information input from the input step.

Images