WO2024105733A1

WO2024105733A1 - Nn growth device, information processing device, neural network information production method, and program

Info

Publication number: WO2024105733A1
Application number: PCT/JP2022/042235
Authority: WO
Inventors: 裕子石若
Original assignee: ソフトバンク株式会社
Priority date: 2022-11-14
Filing date: 2022-11-14
Publication date: 2024-05-23

Abstract

[Problem] Conventionally, the simulation of brain growth has not been possible. [Solution] It is possible to simulate brain growth using an NN growth device 1 comprising: an overall score acquisition unit 135 which acquires an overall score that is based on one or more pieces of image feature information from image information and one or more pieces of audio feature information from audio information; a firing node determination unit 136 which determines one or more node identifiers to pair with an initial firing condition that is satisfied by the one or more pieces of image feature information and the one or more pieces of audio feature information, and determines the node identifier of a node that will fire, said node being a node which is linked by edges to the nodes identified by the one or more node identifiers and to which at least one type of information among the one or more pieces of image feature information and the one or more pieces of audio feature information is passed; and a growth unit 138 which carries out a growth process which, on the basis of difference information for the overall score and a reference score, causes the growth of edge information for one or more edges which connect to the nodes identified by one or more node identifiers among the one or more node identifiers that were determined.

Description

NN growth device, information processing device, neural network information production method, and program

The present invention relates to an NN growth device, which is a device that virtually realizes the mechanism of brain growth.

　Conventionally, there have been electroencephalogram signal processing devices that acquire and process electroencephalogram signals that represent the subject's brain waves (see, for example, Patent Document 1 and Patent Document 2).

JP 2016-47239 A JP 2016-52430 A

However, conventional technology was unable to simulate brain growth.

The NN growth device of the first invention includes an NN storage unit in which neural network information having two or more pieces of node information with node identifiers and one or more pieces of edge information having an edge identifier and identifying connections between the nodes is stored, a start point storage unit in which one or more pieces of ignition start point information having one or more node identifiers that identify nodes that ignite without passing through other nodes is stored in correspondence with an initial ignition condition, which is a condition related to one or more types of feature information out of one or more image feature information for image information and one or more sound feature information for sound information, and which is a condition for determining a node that ignites without passing through other nodes, a standard score storage unit that stores standard scores related to positivity and negativity, an information receiving unit that receives image information and sound information, an image feature acquisition unit that uses the image information received by the information receiving unit to acquire one or more pieces of image feature information for the image information, an overall score acquisition unit that acquires an overall score based on the one or more pieces of image feature information and the one or more pieces of sound feature information, and an image score that is a score related to positivity and negativity, using the one or more pieces of image feature information. The NN growing device includes an image score acquisition unit that acquires one or more sound feature information for the sound information using the sound information received by the information receiving unit, a sound feature acquisition unit that acquires one or more sound feature information for the sound information using the one or more sound feature information, a sound score acquisition unit that acquires a sound score, which is a score for positive and negative, using the one or more sound feature information, a total score acquisition unit that acquires a total score using the image score and the sound score, an ignition node determination unit that determines, from a starting point storage unit, one or more node identifiers that are paired with an initial ignition condition in which one or more types of information out of the one or more image feature information and the one or more sound feature information match, and determines the node identifier of a node that is connected to the node identified by each of the one or more node identifiers by an edge and is passed one or more types of information out of the one or more image feature information and the one or more sound feature information, and a growth unit that performs a growth process that is a process of growing edge information of one or more edges that are connected to a node identified by each of the one or more node identifiers determined by the ignition node determination unit based on difference information regarding the difference between the total score and the reference score.

This configuration makes it possible to simulate brain growth.

The NN growing device of the second invention, compared to the first invention, further includes an image score acquisition unit that uses one or more image feature information to acquire an image score, which is a score related to positivity and negativity, a sound feature acquisition unit that uses sound information accepted by the information acceptance unit to acquire one or more sound feature information for the sound information, and a sound score acquisition unit that uses one or more sound feature information to acquire a sound score, which is a score related to positivity and negativity, and the overall score acquisition unit acquires an overall score using the image score and sound score.

This configuration makes it possible to simulate brain growth.

Furthermore, the NN growing device of the third invention is a NN growing device in which, compared to the first or second invention, edge information has a weight, the firing node determination unit determines the node identifier of a node that is connected by an edge and whose edge weight satisfies a weight-related transmission condition, the growing unit acquires difference information relating to the difference between the total score and the reference score, and, if the difference information meets the first edge growing condition, performs a first edge growing process that increases the weight of the edge information of one or more edges.

This configuration makes it possible to simulate brain growth.

Furthermore, in the NN growth device of the fourth invention, compared to the first or second invention, the edge information may have edge position information indicating the position of the end of the edge, and the growth unit acquires difference information regarding the difference between the total score and the reference score, and when the difference information matches the second edge growth condition, the NN growth device performs a second edge growth process that changes the edge position information contained in the edge information of one or more edges and extends the length of the edge.

This configuration makes it possible to simulate brain growth.

The NN growing device of the fifth invention, compared to the fourth invention, further comprises a goal storage unit in which goal information identifying goals corresponding to two or more states including positive and negative is stored, and a state determination unit that uses difference information acquired by the growth unit to determine one state from two or more states including positive and negative, the goal information having goal position information identifying the position of the goal, or goal direction information indicating the direction of the goal, and the growth unit is a NN growing device that performs a second edge growth process in which it acquires goal information that pairs with the one state determined by the state determination unit, changes the edge position information held by the edge information of each of one or more edges, acquires new edge position information in the direction indicated by the goal information, and accumulates the new edge position information. With this configuration, brain growth can be simulated.

The NN growth device of the sixth invention is a NN growth device according to any one of the first to fifth inventions, further comprising a reference score change unit that changes the reference score in the reference score storage unit based on the total score acquired by the total score acquisition unit.

This configuration makes it possible to simulate brain growth.

The NN growth device of the seventh invention is different from the sixth invention in that the reference score change unit changes the reference score in the reference score storage unit only when the total score meets the score change condition.

This configuration makes it possible to simulate brain growth.

The NN growing device of the eighth invention is a NN growing device according to any one of the first to seventh inventions, in which the standard score storage unit stores standard scores paired with one or more firing patterns using one or more node identifiers, and the growing unit obtains standard scores paired with firing patterns corresponding to one or more node identifiers determined by the firing node determination unit from the standard score storage unit, obtains difference information regarding the difference between the total score and the standard score, and performs a growth process based on the difference information.

This configuration makes it possible to simulate brain growth.

The NN growing device of the ninth invention is a NN growing device according to any one of the first to eighth inventions, in which the firing node determination unit determines whether one or more pieces of feature information passed from one or more other nodes connected by an edge satisfy a firing condition related to one or more pieces of feature information, and determines the node identifier of the node determined to satisfy the firing condition.

This configuration makes it possible to simulate brain growth.

The NN growing device of the tenth invention is different from the first invention in that the node information has node position information that identifies the node position, and further includes a goal storage unit in which goal information that identifies goals corresponding to two or more states including positive and negative is stored, and a state determination unit that uses difference information acquired by the growing unit to determine one state from two or more states including positive and negative, and the growing unit acquires difference information regarding the difference between the total score and the reference score, and based on the difference information, performs edge generation processing to generate and accumulate edge information for edges extending from nodes identified by one or more node identifiers among the one or more node identifiers determined by the firing node determination unit in the direction indicated by the goal information that pairs with the one state determined by the state determination unit.

This configuration makes it possible to simulate brain growth.

The NN growing device of the eleventh invention is a NN growing device in which, compared to the tenth invention, the firing node determination unit stores frequency information relating to the frequency of the determined node identifier in association with the node identifier, and the growing unit performs edge generation processing on the node identified by the node identifier corresponding to the frequency information that meets the edge generation condition.

This configuration makes it possible to simulate brain growth.

The NN growth device of the twelfth invention is an NN growth device according to any one of the first to eleventh inventions, in which the nodes are somas, the edges have AXONs and dendrites, and the edge information has AXON information having an AXON identifier and AXON position information indicating the position of the AXON, and dendrites information having a dendrites identifier and dendrites position information indicating the position of the dendrites.

This configuration makes it possible to simulate brain growth.

The information processing device of the thirteenth invention is an information processing device that includes an NN storage unit in which neural network information accumulated by the NN growing device is stored, an information receiving unit that receives received information that is one or more types of information from image information or sound information, a feature acquisition unit that acquires one or more feature information for the received information received by the information receiving unit, an information transmission unit that determines the node identifier of the node that will ignite, which is a node identifier corresponding to each of the one or more feature information acquired by the feature acquisition unit, from a start point storage unit in which one or more ignition start point information is stored, the start point information having an information identifier that identifies the feature information of the received information and one or more node identifiers that identify nodes that will ignite without passing through other nodes when the feature information is accepted, and is connected by an edge to the nodes identified by the one or more node identifiers, and is a node to which the feature information is passed, and determines the node identifier of the node that will ignite, an ignition pattern acquisition unit that acquires an ignition pattern using the one or more node identifiers determined by the information transmission unit, an output information acquisition unit that acquires output information corresponding to the ignition pattern acquired by the ignition pattern acquisition unit, and an information output unit that outputs the output information.

This configuration makes it possible to simulate brain function using a model of a fully grown brain.

The information processing device of the fourteenth invention is an information processing device that further includes a temperature receiving unit that receives temperature information, and the information transmission unit performs an information transmission process in which the fired node passes characteristic information corresponding to the fired node to the next node that will fire, and changes the processing time for performing the information transmission process according to the temperature information received by the temperature receiving unit.

The neural network growth device of the present invention can simulate brain growth.

Block diagram of NN growth device 1 according to the first embodiment. A flowchart explaining an example of the operation of the NN growing device 1. A flowchart illustrating an example of the score acquisition process A flowchart illustrating an example of the growth process A flowchart illustrating an example of the initial firing node determination process. A flowchart illustrating an example of the ignition transmission process. A flowchart illustrating an example of the ignition determination process. A flowchart illustrating an example of the information acquisition process A flowchart illustrating an example of the single growth process. A flowchart illustrating an example of the edge growth process. A flowchart for explaining an example of the edge extension process. A flowchart illustrating an example of the edge generation process. A flowchart illustrating an example of the edge information generation process. A flowchart for explaining an example of the node generation process A flowchart illustrating an example of the node information generation process A flowchart illustrating an example of the criterion change process. Block diagram of information processing device 2 according to embodiment 2. A flowchart illustrating an example of the operation of the information processing device 2. A flowchart illustrating an example of the information transmission process. A flowchart illustrating an example of a homogeneous transfer process. A flowchart illustrating an example of the ignition determination process. Block diagram of the information processing device 3 Overview of the computer system according to the above embodiment Block diagram of the computer system

Below, embodiments of the NN growth device and the like will be described with reference to the drawings. Note that components with the same reference numerals in the embodiments perform similar operations, and therefore repeated explanations may be omitted.

(Embodiment 1)
In this embodiment, a NN growing device is described that receives image information and sound information, obtains a total score using image feature information obtained from the image information and sound feature information obtained from the sound information, and performs a growth process based on the difference between the total score and a reference score. The growth process here is, for example, an edge growth process, an edge generation process, or a node generation process.

Edge growth processing is processing that grows edges that make up a neural network (hereinafter referred to as "NN" where appropriate). Edge growth processing includes, for example, a first edge growth processing that increases the weight of an edge, and a second edge growth processing that increases the length of an edge. Edge generation processing is processing that generates new edges. Node generation processing is processing that generates nodes that make up a NN.

Note that it is preferable that the neural network used here is a spiking neural network. However, the neural network may be another type of neural network, such as a deep neural network. In other words, the type of neural network is not important.

In this embodiment, for example, when the difference between the total score and the reference score is small, the weight of the edge is increased. Also, in this embodiment, for example, when the difference between the total score and the reference score is large, the length of the edge is increased. Also, in this embodiment, when the length of the edge is increased, for example, the edge is grown in a direction according to the acquired state (for example, positive or negative).

In addition, in this embodiment, a case will be described in which an image score based on image feature information and a sound score based on sound feature information are obtained, and a total score is obtained using the image score and the sound score.

In addition, in this embodiment, a NN growth device in which the reference score changes dynamically will be described. Note that, for example, the reference score changes only when the total score meets a score change condition.

In addition, in this embodiment, we will explain a neural network growing device that stores standard scores according to firing patterns.

In this embodiment, we will also explain a neural network growth device that fires only nodes that meet the firing conditions.

Furthermore, in this embodiment, we will explain the NN growth device in the case where the nodes are somas and the edges have axons and dendrites.

In this specification, information X corresponding to information Y means that information Y can be obtained from information X, or information X can be obtained from information Y, and the method of association is not important. Information X and information Y may be linked, may exist in the same buffer, information X may be included in information Y, or information Y may be included in information X, etc.

FIG. 1 is a block diagram of the NN growing device 1 in this embodiment. The NN growing device 1 includes a storage unit 11, a reception unit 12, a processing unit 13, and an output unit 14. The storage unit 11 includes a starting point storage unit 111, a goal storage unit 112, a reference score storage unit 113, and an NN storage unit 114. The reception unit 12 includes an information reception unit 121. The processing unit 13 includes an image feature acquisition unit 131, a sound feature acquisition unit 132, an image score acquisition unit 133, a sound score acquisition unit 134, a total score acquisition unit 135, an ignition node determination unit 136, a state determination unit 137, a growth unit 138, and a reference score change unit 139.

Various types of information are stored in the storage unit 11 that constitutes the NN growth device 1. The various types of information include, for example, firing start information (described later), goal information (described later), a reference score (described later), a neural network (NN), one or more pieces of glial cell information (described later), one or more pieces of connection information (described later), one or more pieces of firing information (described later), state determination information (described later), and various conditions (described later).

The start point storage unit 111 stores one or more pieces of ignition start point information. Typically, two or more pieces of ignition start point information are stored in the start point storage unit 111. The ignition start point information corresponds to the initial ignition condition.

The ignition start point information is information that specifies the node that will ignite in the first stage when information is accepted. The accepted information is, for example, one or two types of information from among image information and sound information. A node that ignites in the first stage is a node that ignites without going through other nodes. The ignition start point information has, for example, an initial ignition condition and one or more node identifiers.

The initial firing condition is a condition under which a node fires in the first stage. The initial firing condition is a condition related to one or more pieces of feature information. The one or more pieces of feature information are one or more types of information selected from the group consisting of one or more image feature information for image information and one or more sound feature information for sound information. The initial firing condition may be, for example, a condition related to one or more information identifiers. The initial firing condition has, for example, an information identifier. The initial firing condition is, for example, a condition related to an information identifier and an amount of information.

The information identifier is information that identifies feature information. The information identifier is, for example, information that identifies image feature information. The information identifier is, for example, information that specifies the type of image feature information. The information identifier is, for example, information that identifies sound feature information. The information identifier is, for example, "R", "G", or "B". "R" is information that indicates the color red, "G" is information that indicates the color green, and "B" is information that indicates the color blue. Note that image feature information is feature information of image information. The information identifier is, for example, information that indicates a specific frequency or a specific frequency range.

The initial firing condition is, for example, "<information identifier>R <condition> amount of information>=150." The initial firing condition is, for example, "Amount of information of "R" >=150," "(Amount of information of "R" >=150) & (Amount of information of "G" >=80)," "(Amount of information of "R" >=150) & (Amount of information of "G" >=80) & (Amount of information of "B" >=120)," "<information identifier>Frequency=Fa <condition> amount of information>=100," "<information identifier>Fa<=Frequency<Fb <condition> amount of information>=100," "<information identifier>R <condition> amount of information>=150 & <information identifier>Frequency=Fa <condition> amount of information>=180." Note that the frequencies Fa and Fb are values that indicate specific frequencies.

A node identifier is information that identifies a node that constitutes a NN. A node identifier is, for example, a node ID or a node name. A node may also be called a soma. A node identifier may also be called a soma identifier.

Feature information here refers to the feature amount of image information or sound information. Feature information is, for example, an information identifier, or an information identifier and an amount of information. The amount of information is information that indicates the size of information identified by a paired information identifier. Examples of feature information are "<information identifier> R <amount of information> 150", "<information identifier> frequency = Fa <amount of information> 100", and "<information identifier> Fa <= frequency < Fb <amount of information> = 150".

The goal storage unit 112 stores one or more pieces of goal information. Usually, the goal storage unit 112 stores two or more pieces of goal information.

Goal information is information that specifies the goal to which the nodes or edges that make up the NN will grow. Goal information is information that specifies a goal that corresponds to one of two or more states. The state is, for example, an emotion or an internal state of the brain. The state is, for example, positive or negative. It is preferable that the type of state is either positive or negative. However, there may be three or more types of states. When there are three or more types of states, each state is, for example, information indicating one of two or more degrees of positivity or information indicating one of two or more degrees of negativity, or positive, negative, and neutral.

Goal information can be said to be information that specifies the position where a node or edge grows. Goal information has goal position information or goal direction information. Goal information corresponds to a state identifier. Goal position information is information that specifies the position of the goal. Goal direction information indicates the direction of the goal. A position is a position in a virtual space of two or more dimensions. Goal information is, for example, position information. Position information is, for example, three-dimensional coordinate values (x, y, z) or two-dimensional coordinate values (x, y) or a four-dimensional quaternion (x, y, x, w). Furthermore, the fact that goal information corresponds to a state identifier means that the goal information corresponding to the determined state identifier is used for the growth of the NN. Note that a state identifier is information that identifies a state. State identifiers are, for example, "positive" and "negative."

The standard score storage unit 113 stores one or more standard scores. Each of the one or more standard scores is paired with, for example, an ignition pattern. However, when only one standard score is stored in the standard score storage unit 113, the standard score is not paired with an ignition pattern. The standard score storage unit 113 may store one default standard score.

The base score is the standard score used when determining positive, negative, and other states. The score is information that indicates the degree of a state, such as positive or negative. The base score can also be called an expected value.

A firing pattern is a pattern of firing one or more nodes. A firing pattern has one or more node identifiers of the nodes that fire.

The NN storage unit 114 stores neural network information (hereinafter referred to as "NN information" where appropriate). NN information can be said to be information that mimics the brain. NN information has two or more pieces of node information and one or two or more pieces of edge information. NN information is sometimes referred to as NN.

Node information is information about the nodes that make up the NN. The node information has a node identifier. The node information has, for example, node position information, firing conditions, firing probability information, and number of times information. In addition, it is preferable that the node information has required energy amount information that indicates the amount of energy required for firing.

The node position information is the position information of a node. As described above, the position information is, for example, three-dimensional coordinate values (x, y, z), two-dimensional coordinate values (x, y), or a four-dimensional quaternion (x, y, x, w).

The firing condition is the condition under which a node fires. The firing condition usually has one or more pieces of characteristic information. The characteristic information may have an information identifier that identifies the information and an amount of information that indicates the size of the information, or may be only the amount of information that indicates the size of the information. The amount of information is, for example, a numerical value greater than 0. Examples of firing conditions are "R>=180", "R>=150 & G>=100", "G<=50 & B<=30", "R>=250 & G>=250 & B>=250", "<information identifier> frequency = Fa <condition> amount of information>=80", "<information identifier> Fa <= frequency <Fb <condition> amount of information>=150", "<information identifier> R <condition> amount of information>=150 & <information identifier> frequency = Fa <condition> amount of information>=180", etc.

Firing probability information is information about the probability of firing. The firing probability information may be the firing probability itself, or a value obtained by converting the firing probability using a function or the like. It is preferable that the firing probability information is referenced, and the node may or may not fire at the probability indicated by the firing probability information, even if the feature information is the same.

The number of occurrences information is information based on the number of occurrences of firing. The number of occurrences information is, for example, the number of occurrences of firing, or the firing frequency (firing rate).

Edge information is information about edges that make up a NN. Edge information is information that specifies connections between nodes. Edge information usually has an edge identifier. Edge information has, for example, a node identifier for each of two nodes that an edge connects. Edge information has, for example, a node identifier for one connecting node. When edge information has only one node identifier, the edge is an edge in the growth process before connecting two nodes. Edge information has, for example, edge position information. Edge position information is information that specifies the position of the end point of an edge. Edge information has, for example, a weight. The larger the edge weight, the easier it is for feature information to be transmitted via the edge. The edge weight may be a parameter of firing probability information. In other words, the larger the edge weight, the higher the firing probability. In addition, it is preferable that the edge information has retained energy amount information that indicates the amount of energy held by the edge.

Edge information includes, for example, dendrites information and AXON information. In such a case, an edge includes dendrites and AXON. Note that an edge may be considered to be a single line, or a line that branches into two or more branches. An edge may also be called a synapse.

An edge identifier is information that identifies an edge. For example, an edge identifier is an edge ID or an edge name.

Dendrites information is information about DENDRITES. DENDRITES are also called dendrites and are part of nerve cells. They are multiple projections that branch out from the cell body like the branches of a tree in order for nerve cells to receive external stimuli and information sent from the axons (AXONs) of other nerve cells. In this case, DENDRITES are the elements that make up an edge. DENDRITES information has a DENDRITES identifier and DENDRITES position information.

The DENDRITES identifier is information that identifies the DENDRITES. For example, the DENDRITES identifier is the DENDRITES ID or the DENDRITES name.

DENDRITES position information is position information that indicates the position of DENDRITES. DENDRITES position information is information that specifies the position of DENDRITES, and is, for example, one or more three-dimensional coordinate values (x, y, z), or one or more two-dimensional coordinate values (x, y). When DENDRITES position information has two or more coordinate values, DENDRITES is a line connecting each point of the two or more coordinate values.

It is also preferable that the dendrites information includes information on the amount of stored energy that indicates the amount of energy that the dendrites hold. It is also preferable that the dendrites information includes information on the amount of required energy that is required to transmit information using the dendrites.

Axon information is information about an axon. An axon is also called an axon, and is a protruding structure that extends from the cell body and is responsible for outputting signals in a nerve cell. In this case, an axon is an element that constitutes an edge. Axon information has an axon identifier and axon position information.

The AXON identifier is information that identifies the AXON. For example, the AXON ID or the AXON name.

AXON position information is position information that indicates the position of AXON. AXON position information is information that specifies the position of AXON, and is, for example, one or more three-dimensional coordinate values (x, y, z), or one or more two-dimensional coordinate values (x, y). When AXON position information has two or more coordinate values, AXON is a line connecting each point of the two or more coordinate values.

It is also preferable that the AXON information includes information on the amount of stored energy that indicates the amount of energy that the AXON holds. It is also preferable that the AXON information includes information on the amount of required energy that is required to transmit information using the AXON.

Note that dendrites and AXON may branch out. When dendrites and AXON branch out, the position information of each may be expressed by three or more coordinate values. However, there is no restriction on the method of expressing dendrites position information and AXON position information.

The glial cell information stored in the storage unit 11 is information about glial cells. Note that the glial cell information does not have to be present in the storage unit 11.

Here, glial cells, also known as neuroglial cells, are a general term for non-neuronal cells that make up the nervous system. Glial cells are a glue- or cement-like substance that fills the spaces between neurons.

It is preferable that the glial cell information has a glial cell identifier for identifying the glial cell. The glial cell information has, for example, a node identifier for identifying a node that assists the binding, or an edge identifier for identifying an edge that assists the binding. The glial cell information has, for example, an axon identifier for an axon for which the glial cell assists the binding, or a dendrites identifier for dendrites for which the glial cell assists the binding. The glial cell information may also have a glial cell type identifier for identifying the type of the glial cell. The type of glial cell is, for example, oligodendrocites (hereinafter referred to as "oligo" as appropriate) or astrocites. Note that oligos are cells that can connect to axons. Astrocites are cells that can connect to somas or dendrites. It is also preferable that the glial cell information has glial cell position information. The glial cell position information is position information that specifies the position of the glial cell. In particular, it is preferable that the oligo glial cell information includes glial cell position information. The glial cell information may also include hand length information indicating the length of each of one or more hands. The glial cell information may also include hand number information indicating the number of hands emerging from the glial cell. And, typically, when the total length of the glial cell calculated from the hand length information of each hand reaches a threshold value, it is preferable that the glial cell does not grow any further.

The connection information stored in the storage unit 11 is information that specifies connections between two or more nodes. The connection information may be information that specifies a connection between an AXON of one node and dendrites of another node. Such information is also information that specifies a connection between nodes. The connection information may be information that specifies a connection between a synapse and a spine. Such information is also information that specifies a connection between nodes. The connection information has, for example, identifiers of two nodes to be connected. The connection information also has, for example, an AXON identifier of an AXON and a dendrites identifier of a dendrite that is connected to the AXON. The connection information also has, for example, a synapse identifier of a synapse and a spine identifier of a spine that can transmit information between the synapse. The connection information may have information transmission probability information. The information transmission probability information is information regarding the probability of information transmission between one node and another node. The information transmission probability information may be information regarding the probability of information transmission between an AXON and a dendrite. In such a case, the information transmission probability information is information about the probability of information transmission between one node and another node. The information transmission probability information may also be information about the probability of information transmission between a synapse and a spine. In such a case, the information transmission probability information is information about the probability of information transmission between one node and another node. Note that the connection direction between the nodes is usually unidirectional.

The binding information may be information indicating a binding between a node and an AXON. In such a case, the binding information has a node identifier and an AXON identifier. The binding information may also be information indicating a binding between a node and Dendrites. In such a case, the binding information has a node identifier and a Dendrites identifier.

The binding information may be information that specifies the binding between a glial cell and an AXON or dendrites. In such a case, the binding information may include, for example, a glial cell identifier that identifies glial cell information, and an AXON identifier. The binding information may include, for example, a glial cell identifier and a dendrites identifier.

Note that connection information specifying the connections between the elements (nodes, edges, axons, dendrites, glial cells, synapses, or spines) that make up the NN may be stored in the NN storage unit 114. In other words, connection information specifying the connections between the elements that make up the NN may be included in the information of each element.

Ignition information is information about the result of an ignition. The ignition information has a node identifier that identifies the node that ignited. The ignition information may usually have timer information that indicates the time of ignition. The timer information may be information that indicates a relative time, or time information that indicates an absolute time. The ignition information may be automatically deleted by the processing unit 13 after a certain time has passed since it was accumulated.

The state determination information is information for determining a state using one or more types of information from among image information and sound information. The state determination information is, for example, a condition based on difference information, which will be described later. The state determination information is, for example, two or more sets having an image condition, which is a condition related to image information, and a state identifier. The state determination information is, for example, two or more sets having a sound condition, which is a condition related to sound information, and a state identifier. The state determination information is, for example, two or more sets having an image sound condition, which is a condition related to image information and sound information, and a state identifier.

The state determination information is, for example, "difference information >= threshold A <state identifier> state P", "difference information <threshold B <state identifier> state N", and the state determination information is, for example, "<image condition> amount of information of R >= 150 <state identifier> state P", "<image condition> (amount of information of "R" <= 10) & (amount of information of "G" <= 80) <state identifier> state N", "<sound condition> amount of information of information identifier "frequency = X" >= 50 <state identifier> state P", "<sound condition> amount of information of information identifier "frequency = X" >= 80 <state identifier> state N", "<sound condition> amount of information of information identifier "frequency = Y" >= 50 & amount of information of information identifier "frequency = Z" >= 90 <state identifier> state P", "<image sound condition> amount of information of R >= 150 & amount of information of information identifier "frequency = Y" >= 50 <state identifier> state P". Note that state P has a state identifier "positive" and state N has a state identifier "negative."

The reception unit 12 receives various types of information. Examples of the various types of information include image information and sound information.

Here, reception is a concept that includes the reception of information input from input devices such as cameras, microphones, keyboards, mice, and touch panels, the reception of information transmitted via wired or wireless communication lines, and the reception of information read from recording media such as optical disks, magnetic disks, and semiconductor memories.

The information receiving unit 121 receives information. For example, the information receiving unit 121 receives one or two types of information, image information and sound information. For example, the information receiving unit 121 receives image information and sound information at the same time. However, for example, the information receiving unit 121 may receive image information and sound information with some delay. Furthermore, the image information is a still image or a video. The sound information is, for example, audio data or music data, but the type is not important as long as it is sound information.

The information receiving unit 121, for example, acquires image information captured by a camera. The information receiving unit 121, for example, acquires sound information acquired by a microphone. In other words, "acceptance" refers to the acceptance of information acquired by a device such as a microphone or camera, but may also be a concept that includes the reception of information transmitted via a wired or wireless communication line, and the acceptance of information read from a recording medium such as an optical disk, a magnetic disk, or a semiconductor memory.

The processing unit 13 performs various types of processing. For example, various types of processing are performed by the image feature acquisition unit 131, the sound feature acquisition unit 132, the growth unit 138, etc.

The image feature acquisition unit 131 uses the image information accepted by the information acceptance unit 121 to acquire one or more pieces of image feature information for the image information.

The sound feature acquisition unit 132 uses the sound information accepted by the information acceptance unit 121 to acquire one or more pieces of sound feature information for the sound information.

The image score acquisition unit 133 acquires an image score using one or more pieces of image feature information acquired by the image feature acquisition unit 131.

An image score is a score for the received image information. An image score is typically a score related to positivity and negativity. For example, an image score is information indicating the degree of positivity or information indicating the degree of negativity.

The image score acquisition unit 133 acquires the image score by, for example, one of the following three methods.
(1) Method using an arithmetic formula

The image score acquisition unit 133 substitutes the amount of information possessed by one or more pieces of image feature information acquired by the image feature acquisition unit 131 into an image calculation formula, executes the image calculation formula, and calculates an image score. Note that the image calculation formula is a formula in which the amount of information possessed by one or more pieces of image feature information is a parameter. The image calculation formula is stored in the storage unit 11.
(2) Using a correspondence table

The image score acquisition unit 133 acquires, from the image correspondence table, an image score that is paired with a vector that is most similar to a vector whose elements are the amount of information possessed by one or more pieces of image feature information acquired by the image feature acquisition unit 131. The image correspondence table is a table that indicates the correspondence between a set of one or more pieces of image feature information and an image score. The image correspondence table has two or more pieces of image correspondence information. The image correspondence information is information that indicates the correspondence between a vector whose elements are the amount of information possessed by one or more pieces of image feature information and an image score. The image correspondence information is, for example, a pair of a vector and an image score.
(3) Machine learning method

The image score acquisition unit 133 provides a vector whose elements are the amount of information contained in one or more pieces of image feature information acquired by the image feature acquisition unit 131 and an image learning model to a machine learning prediction module, executes the prediction module, and acquires an image score.

The image learning model is information obtained by providing two or more pieces of teacher data having vectors whose elements are the amount of information contained in one or more pieces of image feature information and image scores to a machine learning learning module and executing the learning module. The image learning model is stored in the storage unit 11.

The sound score acquisition unit 134 acquires a sound score using one or more pieces of sound feature information acquired by the sound feature acquisition unit 132.

A sound score is a score for received sound information. A sound score is typically a positive and negative score. A sound score is, for example, information that indicates the degree of positivity or information that indicates the degree of negativity.

The sound score acquisition unit 134 acquires the sound score by, for example, one of the following three methods.
(1) Method using an arithmetic formula

The sound score acquisition unit 134 assigns the amount of information possessed by each of the one or more pieces of sound feature information acquired by the sound feature acquisition unit 132 to a sound calculation formula, executes the sound calculation formula, and calculates a sound score. Note that the sound calculation formula is a formula in which the amount of information possessed by each of the one or more pieces of sound feature information is a parameter. The sound calculation formula is stored in the storage unit 11.
(2) Using a correspondence table

The sound score acquisition unit 134 acquires from the sound correspondence table a sound score that is paired with a vector that is most similar to a vector whose elements are the amount of information possessed by one or more pieces of sound feature information acquired by the sound feature acquisition unit 132. The sound correspondence table is a table that shows the correspondence between a set of one or more pieces of sound feature information and a sound score. The sound correspondence table has two or more pieces of sound correspondence information. The sound correspondence information is information that shows the correspondence between a vector whose elements are the amount of information possessed by one or more pieces of sound feature information and a sound score. The sound correspondence information is, for example, a pair of a vector and a sound score.
(3) Machine learning method

The sound score acquisition unit 134 provides a vector whose elements are the amount of information contained in one or more pieces of sound feature information acquired by the sound feature acquisition unit 132 and a sound learning model to a machine learning prediction module, executes the prediction module, and acquires a sound score.

The sound learning model is information obtained by providing two or more pieces of teacher data having a vector whose elements are the amount of information contained in one or more pieces of sound feature information and a sound score to a machine learning learning module and executing the learning module. The sound learning model is stored in the storage unit 11.

Learning models, such as the image learning model, sound learning model, and comprehensive learning model described below, are information constructed by the learning process of machine learning, and are information used in the prediction process of machine learning. A learning model may also be called a learner, classifier, classification model, etc. The above machine learning algorithms may be deep learning, random forest, decision tree, SVR, etc. In addition, for machine learning, various machine learning functions such as the TensorFlow library, the random forest module of the R language, fastText, TinySVM, and various existing libraries can be used.

The overall score acquisition unit 135 acquires an overall score based on one or more pieces of image feature information and one or more pieces of sound feature information.

The overall score acquisition unit 135 acquires an overall score, for example, using the image score and the sound score. The overall score acquisition unit 135 acquires a larger overall score the larger the image score value. The overall score acquisition unit 135 acquires a larger overall score the larger the sound score value. The overall score acquisition unit 135 acquires an overall score, for example, by an increasing function with the image score and the sound score as parameters. The overall score acquisition unit 135 acquires an overall score that is, for example, the sum of the image score and the sound score. The overall score acquisition unit 135 acquires an overall score that is, for example, the average value of the image score and the sound score. The overall score acquisition unit 135 acquires an overall score that is, for example, the weighted average value of the image score and the sound score.

The overall score acquiring unit 135 may acquire the overall score without using the image score or the sound score. In such a case, the overall score acquiring unit 135 acquires the sound score by, for example, one of the following three methods.
(1) Method using an arithmetic formula

The overall score acquisition unit 135 substitutes one or more of the amounts of information contained in the one or more pieces of image feature information acquired by the image feature acquisition unit 131 and the amount of information contained in the one or more pieces of sound feature information acquired by the sound feature acquisition unit 132 into an overall calculation formula, executes the overall calculation formula, and calculates an overall score. Note that the overall calculation formula is a formula in which the amount of information contained in each of the one or more pieces of feature information is a parameter. The overall calculation formula is stored in the storage unit 11.
(2) Using a correspondence table

The overall score acquisition unit 135 acquires from the overall correspondence table an overall score paired with a vector that is most similar to a vector whose elements are the amounts of information contained in one or more pieces of image feature information acquired by the image feature acquisition unit 131 and one or more pieces of sound feature information acquired by the sound feature acquisition unit 132. The overall correspondence table is a table showing the correspondence between a set of one or more pieces of feature information and an overall score. The overall correspondence table has two or more pieces of overall correspondence information. The overall correspondence information is information showing the correspondence between a vector whose elements are the amounts of information contained in one or more pieces of feature information and an overall score. The overall correspondence information is, for example, a pair of a vector and an overall score.
(3) Machine learning method

The overall score acquisition unit 135 provides a vector whose elements are the amount of information contained in one or more pieces of image feature information acquired by the image feature acquisition unit 131 and one or more pieces of sound feature information acquired by the sound feature acquisition unit 132, and an overall learning model to a machine learning prediction module, executes the prediction module, and acquires an overall score.

The comprehensive learning model is information obtained by providing two or more pieces of teacher data having a vector whose elements are the amount of information contained in one or more pieces of feature information and an overall score to a machine learning learning module and executing the learning module. The comprehensive learning model is stored in the storage unit 11.

The ignition node determination unit 136 determines from the starting point storage unit 111 one or more node identifiers that are paired with an initial ignition condition in which one or two types of information out of one or more image feature information and one or more sound feature information match.

Next, the ignition node determination unit 136 determines the node identifier of a node that is connected by an edge to each ignition node identified by the one or more node identifiers and that is to be ignited, and that receives one or more pieces of feature information from the ignition node. Note that the node identifier of the ignition node is appropriately referred to as the ignition node identifier. Note that the one or more pieces of feature information are one or two types of information from among one or more pieces of image feature information and one or more pieces of sound feature information.

The firing node determination unit 136 determines the node identifier of a node that is connected by an edge, for example, and the weight of the edge satisfies a transmission condition. A transmission condition is a condition for transmitting feature information from one node to another node that is connected to the node by an edge. The transmission condition is a condition related to the weight. For example, the transmission condition is "weight >= threshold" or "weight > threshold".

The firing node determination unit 136, for example, determines whether one or more pieces of feature information passed from one or more other nodes connected by an edge satisfy a firing condition related to one or more pieces of feature information, and determines the node identifier of the node determined to satisfy the firing condition. Note that the firing condition is a condition related to one or more pieces of feature information.

It is preferable that the firing node determination unit 136 stores the number of times information regarding the number of times of the determined node identifier in association with the node identifier. In other words, it is preferable that the firing node determination unit 136 increases the number of times information of the fired node.

The state determination unit 137 uses the difference information to determine one state from two or more states including positive and negative. Note that the two or more states are, for example, "positive" and "negative", or "positive", "negative", and "neutral".

Difference information is information about the difference between the total score and the standard score. Examples of difference information are "total score - standard score," "absolute value of the difference between the total score and the standard score," "total score/standard score," and "standard score/total score."

The state determination unit 137, for example, obtains difference information indicating the difference between the total score obtained by the total score acquisition unit 135 and the reference score. Next, the state determination unit 137 obtains a state identifier corresponding to the difference information. Note that the difference information may be information obtained by the growth unit 138, which will be described later.

The state determination unit 137, for example, obtains the total score obtained by the total score acquisition unit 135. The state determination unit 137 also determines an ignition pattern that matches one or more node identifiers determined by the ignition node determination unit 136 from the standard score storage unit 113, and obtains a standard score that pairs with the ignition pattern from the standard score storage unit 113. Next, the state determination unit 137 obtains difference information between the total score and the standard score. Next, the state determination unit 137 obtains a state identifier that corresponds to the difference information.

The status determination unit 137 determines the status to be "positive" if, for example, "total score - standard score (difference information) >= threshold A." The status determination unit 137 determines the status to be "negative" if, for example, "total score - standard score < threshold B." The status determination unit 137 determines the status to be "neutral" if, for example, "threshold B <= total score - standard score < threshold A."

The growth unit 138 performs a growth process. The growth unit 138 performs a growth process of a NN that mimics the brain. The growth unit 138 performs a growth process of the NN using the received image information and sound information. The growth unit 138 performs a process of growing the nodes or edges, or the nodes and edges, that constitute the NN, using one or more pieces of image feature information acquired from the received image information and one or more pieces of sound feature information acquired from the sound information.

More specifically, the growth unit 138 acquires difference information regarding the difference between the overall score and the reference score, and performs growth processing based on the difference information. The growth unit 138 performs growth processing when the difference information matches the growth conditions.

The growth unit 138, for example, obtains from the standard score storage unit 113 a standard score that is paired with an ignition pattern corresponding to one or more node identifiers determined by the ignition node determination unit 136. Next, the growth unit 138 obtains difference information regarding the difference between the obtained total score and the standard score, and performs a growth process based on the difference information.

The growth process is, for example, an edge growth process, an edge creation process, and a node creation process. The edge growth process is, for example, a first edge growth process and a second edge growth process. Each process will be described below.
(1) Edge growth process

The growing unit 138 performs, for example, an edge growing process. The edge growing process is a process of growing an edge. The edge growing process is, for example, a process of increasing the weight of an edge. Increasing the weight of an edge means making the edge thicker. The edge growing process is, for example, a process of increasing the length of an edge. The edge growing process may be a process of connecting the edge from the node from which the edge originates to another node.
(1-1) First edge growth treatment

The growth unit 138 acquires difference information regarding the difference between the total score and the reference score, and when the difference information matches the first edge growth condition, performs a first edge growth process that increases the weight of the edge information of one or more edges connected to the fired node.

The first edge growth condition is a condition for changing the weight of an edge. For example, the first edge growth condition is "difference information<=threshold" or "difference information<threshold". In other words, the edge weight is usually increased when the difference between the total score and the reference score is small. The first edge growth condition may have degree information indicating the degree of weight change. For example, the first edge growth condition is "if (difference information<=threshold) then weight+5" or "if (difference information<threshold) then weight+3". Note that "weight+5" indicates that the weight is increased by "3".
(1-2) Second edge growth treatment

The growth unit 138 acquires difference information regarding the difference between the total score and the reference score, and if the difference information matches the second edge growth condition, performs a second edge growth process that changes the edge position information contained in the edge information of one or more edges connected to nodes identified by one or more node identifiers determined by the ignition node determination unit 136, and extends the length of the edges.

The second edge growth condition is a condition for extending the length of an edge. The second edge growth condition is, for example, a condition based on difference information or a condition based on count information. The second edge growth condition is, for example, "difference information >= threshold", "difference information > threshold", "count information >= threshold", or "count information > threshold". Usually, when the difference between the total score and the reference score is large, the length of the edge that is connected to the fired node and is not connected to other nodes is extended.

The growth unit 138, for example, acquires goal information paired with a state identifier (for example, "positive" or "negative") of one state determined by the state determination unit 137. Next, the growth unit 138 performs a second edge growth process, for example, acquiring edge position information contained in the edge information of each of one or more edges, acquiring new edge position information in the direction indicated by the goal information for the edge position information, and accumulating the new edge position information.

The target of the second edge growth process is, for example, an edge connected to one or more nodes determined by the ignition node determination unit 136. The target of the second edge growth process may be, for example, a portion of edges that meet a specific condition, among the one or more nodes determined by the ignition node determination unit 136. The specific condition is, for example, a condition based on weight. The specific condition is, for example, "weight is equal to or greater than a threshold" or "weight is greater than a threshold."

The second edge growth process may be one or more of a Dendrites growth process (to be described later) and an AXON growth process (to be described later).
(1-3) Dendrites growth treatment

The growth unit 138 performs, for example, a dendrites growth process. The dendrites growth process is a process for growing dendrites. The dendrites growth process may be included in the edge growth process. The process for growing dendrites is usually a process for increasing the length of the dendrites. The process for growing dendrites is a process for acquiring new dendrites position information for the dendrites position information contained in the dendrites information of the dendrites, which sets the position of the end point indicated by the dendrites position information to a position away from the position of the connected node, and storing the new dendrites position information.

The growth unit 138 performs a dendrites growth process, for example, by acquiring and accumulating dendrites information by growing dendrites extending from nodes identified by one or more node identifiers among one or more node identifiers determined by the ignition node determination unit 136 in a direction indicated by goal information paired with a state determined by the state determination unit 137.

It is preferable that one or more of the node identifiers determined by the firing node determination unit 136 are node identifiers of one or more nodes that meet the Dendrites growth condition among the one or more node identifiers determined by the firing node determination unit 136. However, one or more of the node identifiers determined by the firing node determination unit 136 may be all of the node identifiers determined by the firing node determination unit 136.

Note that obtaining dendrites information by growing dendrites extending from a node means changing the dendrites position information contained in the dendrites information to position information in which the end point of the dendrite indicated by the dendrites position information is located farther away from the node.

Dendrites growth conditions are conditions for performing the dendrites growth process. Dendrites growth conditions are, for example, conditions based on difference information or conditions based on number information. Dendrites growth conditions are, for example, "difference information >= threshold", "difference information > threshold", "number information >= threshold", or "number information > threshold". Dendrites growth conditions may be the same as the second edge growth conditions, or may be different.

In addition, the process of growing dendrites extending from a node is a process of obtaining position information in the direction indicated by the goal information from the dendrites position information contained in the dendrites information that pairs with the node identifier that identifies the node, and setting this position information as the dendrites position information.
(1-4) AXON growth treatment

The growth unit 138 performs, for example, an AXON growth process. The AXON growth process is a process for growing an AXON. The AXON growth process may be included in an edge growth process. The AXON growth process is usually a process for increasing the length of an AXON. The AXON growth process is a process for acquiring new AXON position information for the AXON position information contained in the AXON information of the AXON, which is a position that is farther away from the position of the connected node than the position of the end point indicated by the AXON position information, and storing the new AXON position information.

The growth unit 138 performs an AXON growth process, for example, by acquiring and storing AXON information obtained by growing an AXON extending from a node identified by one or more node identifiers among one or more node identifiers determined by the ignition node determination unit 136 in a direction indicated by goal information paired with a state determined by the state determination unit 137.

It is preferable that one or more of the node identifiers determined by the firing node determination unit 136 are node identifiers of one or more nodes that meet the AXON growth condition among the one or more node identifiers determined by the firing node determination unit 136. However, one or more of the node identifiers determined by the firing node determination unit 136 may be all of the node identifiers determined by the firing node determination unit 136.

Note that obtaining AXON information by growing an AXON extending from a node means changing the AXON position information contained in the AXON information to position information in which the end point of the AXON indicated by the AXON position information is located farther away from the node.

The AXON growth conditions are conditions for performing the AXON growth process. The AXON growth conditions are conditions based on, for example, difference information and number information. The AXON growth conditions are, for example, "difference information >= threshold", "difference information > threshold", "number information >= threshold", and "number information > threshold". The AXON growth conditions may be the same as the second edge growth conditions, or may be different.

In addition, the process of growing an AXON extending from a node is a process of obtaining position information in the direction indicated by the goal information from the AXON position information contained in the AXON information that pairs with the node identifier that identifies the node, and treating this position information as the AXON position information.
(2) Edge generation process

The growth unit 138 performs, for example, edge generation processing. The edge generation processing can be said to be processing for generating new edges. The edge generation processing is processing for generating new edge information and storing it in the NN storage unit 114.

The growth unit 138 acquires difference information regarding the difference between the total score and the reference score, and based on the difference information, generates and accumulates edge information for edges extending from nodes identified by one or more of the one or more node identifiers determined by the ignition node determination unit 136 in a direction indicated by the goal information paired with a state determined by the state determination unit 137.

The growth unit 138 performs edge generation processing on a node identified by a node identifier, for example, if the edge generation conditions are met.

The edge generation condition is a condition for generating edge information. The edge generation condition is, for example, a condition based on difference information. The edge generation condition may be a condition based on count information. The edge generation condition is, for example, "difference information >= threshold", "difference information > threshold", "count information >= threshold", or "count information > threshold". The edge generation condition may be common to all nodes of interest, may be different for each node, or may be different for each edge. When the edge generation condition is different for each node, for example, the node information has the edge generation condition. When the edge generation condition is different for each edge, for example, the edge information has the edge generation condition.

The growth unit 138 generates and stores edge information of edges extending from nodes identified by one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 at a position in a direction indicated by goal information that pairs with a state determined by the state determination unit 137.

It is preferable that the one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 are one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 that match the edge generation condition. However, the one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 may be all node identifiers determined by the firing node determination unit 136.

The process of generating edge information is a process of generating edge information of an edge connected to a node identified by a target node identifier. The process of generating edge information is, for example, a process of acquiring a unique edge identifier, and generating edge information having an edge identifier, which is information on an edge connecting a node identified by a target node identifier to another node in a direction indicated by goal information from the node. Such edge information has, for example, an edge identifier and the node identifiers of the two nodes to be connected. The process of generating edge information is, for example, a process of acquiring a unique edge identifier, acquiring edge position information that extends from a node identified by a target node identifier and specifies the position of its end point from the node in the direction indicated by goal information, and generating edge information having the edge position information. Such edge information has, for example, an edge identifier, a node identifier of a target (connecting) node, and edge position information that specifies the position of the end point of the edge.

The edge generation process may be one or more of a dendrites generation process and an AXON generation process, which will be described later. The edge generation process may also include a glial cell generation process, which will be described later.
(2-1) Dendrites generation process

The growth unit 138 performs, for example, a dendrites generation process. The dendrites generation process is a process for generating new dendrites. The dendrites generation process may include a process for generating new dendrites information and storing it in the NN storage unit 114. In other words, the growth unit 138 generates and stores dendrites information for dendrites extending from a node identified by one or more node identifiers among one or more node identifiers determined by the ignition node determination unit 136 in a direction indicated by goal information paired with a state determined by the state determination unit 137.

It is preferable that the one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 are one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 that match the Dendrites generation condition. However, the one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 may be all node identifiers determined by the firing node determination unit 136.

The process of generating dendrites information is a process of generating dendrites information for dendrites connected to a node identified by a target node identifier. The process of generating dendrites information is a process of, for example, acquiring a unique dendrites identifier, acquiring dendrites position information in the direction indicated by the goal information from the node identified by the target node identifier (the node identifier of the node to which the dendrites are connected), and constructing and storing dendrites information having the dendrites identifier and the dendrites position information.

Moreover, the dendrites generation condition is a condition for generating dendrites. The dendrites generation condition is, for example, a condition based on difference information or number information. The dendrites generation condition is, for example, "difference information>=threshold", "difference information>threshold", "number information>=threshold", or "number information>threshold". The dendrites generation condition may be the same as the edge generation condition, or may be different.
(2-2) AXON generation process

The growth unit 138 performs, for example, an AXON generation process. The AXON generation process is a process for generating a new AXON. The AXON generation process may include a process for generating new AXON information and storing it in the NN storage unit 114. In other words, the growth unit 138 generates and stores AXON information for an AXON extending from a node identified by one or more node identifiers among one or more node identifiers determined by the ignition node determination unit 136 in a direction indicated by goal information paired with a state determined by the state determination unit 137.

It is preferable that the one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 are one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 that meet the AXON generation condition. However, the one or more node identifiers among the one or more node identifiers determined by the firing node determination unit 136 may be all node identifiers determined by the firing node determination unit 136.

The process of generating AXON information is a process of generating AXON information of an AXON connected to a node identified by a target node identifier (the node identifier of the node to which the AXON is connected). The process of generating AXON information is, for example, a process of acquiring a unique AXON identifier, acquiring AXON position information in the direction indicated by the goal information from the node identified by the target node identifier, and constructing and storing AXON information having the AXON identifier and the AXON position information.

The AXON generation condition is a condition for generating an AXON. The AXON generation condition is, for example, a condition based on difference information or number information. The AXON generation condition is, for example, "difference information>=threshold", "difference information>threshold", "number information>=threshold", or "number information>threshold". The AXON generation condition may be the same as the edge generation condition, or may be different.
(3) Node generation process

The growth unit 138 performs, for example, a node generation process. The node generation process is a process of generating new node information. In other words, the growth unit 138 performs a node generation process to generate and store node information of a new node in a position in the vicinity of the position indicated by the node position information of a node identified by one or more node identifiers among the one or more node identifiers acquired by the ignition node determination unit 136, which is a position in the direction indicated by the goal information that pairs with one state determined by the state determination unit 137.

The growth unit 138, for example, acquires a new node identifier. The growth unit 138 also acquires new node position information for a position in a direction indicated by goal information that pairs with a state determined by the state determination unit 137 and that is a predetermined distance away from the position indicated by the node position information of the target node (firing node). The growth unit 138 also acquires information (for example, firing condition or firing probability information) contained in the node information of the target node. The growth unit 138 then constructs node information that includes, for example, one or more pieces of information among the new node identifier, the new node position information, and the firing condition or firing probability information, and accumulates the information in the NN storage unit 114. The predetermined distance may be a predetermined distance, or may change dynamically.

It is preferable that each of the one or more node identifiers among the one or more node identifiers is, for example, one or more node identifiers included in the node information that matches the node generation condition among the one or more node identifiers. However, the one or more node identifiers among the one or more node identifiers may be, for example, all of the node identifiers among the one or more node identifiers.

A node generation condition is a condition for generating a node. The node generation condition is, for example, a condition based on difference information. The node generation condition is, for example, "difference information >= threshold" or "difference information >threshold". The node generation condition may be the same as the edge generation condition or may be different. Furthermore, the node generation condition may be common to all nodes of interest or may be different for each node. When the node generation condition is different for each node, for example, the node information has the node generation condition.
(4) Glial cell generation treatment

It is preferable that the growth unit 138 performs the following glial cell generation process. That is, for example, when the amount of energy held by an element that is a node or edge becomes small enough to satisfy a predetermined condition relative to the amount of required energy, the growth unit 138 generates glial cell information that connects to the element. Note that the element may be an axon or a dendrite. That is, when the amount of energy held by an element that is an axon or a dendrite becomes small enough to satisfy a predetermined condition relative to the amount of required energy, the growth unit 138 generates glial cell information that connects to the element.

The predetermined condition is, for example, "retained energy amount < required energy amount", or "retained energy amount <= required energy amount", or "retained energy amount - required energy amount <= threshold value", or "retained energy amount - required energy amount < threshold value".

More specifically, the growth unit 138, for example, compares the amount of retained energy indicated by the retained energy amount information of the information of each element (node information, edge information, AXON information, or Dendrites information) with the amount of required energy indicated by the required energy amount information of the information of each element to determine whether the amount of retained energy is small enough to satisfy a predetermined condition, and if it determines that the amount is small, generates glial cell information having an identifier that identifies the element (node identifier, edge identifier, AXON identifier, or Dendrites identifier) and stores it in the storage unit 11.

The standard score modification unit 139 modifies the standard score in the standard score storage unit 113 based on the total score acquired by the total score acquisition unit 135. For example, the standard score modification unit 139 modifies the standard score to the total score. For example, the standard score modification unit 139 modifies the standard score using an increasing function (for example, an average value, a weighted average value) with the original standard score and the total score as parameters.

It is preferable that the standard score change unit 139 changes the standard score in the standard score storage unit 113 only when the total score matches the score change condition. It is preferable that the score change condition is a condition based on difference information between the total score and the standard score. For example, the score change condition is "difference information >= threshold" or "difference information > threshold". It is preferable that the degree to which the standard score is changed is based on, for example, the difference information or the total score. For example, the degree to which the standard score is changed is "average value of the total score and the standard score" or "weighted average value of the total score and the standard score".

The output unit 14 outputs various information. The various information is, for example, the identifier of the fired node, the NN information in the NN storage unit 114, and the state identifier acquired by the state determination unit 137.

The various types of information are, for example, information that graphically illustrates the NN information. In such a case, the processing unit 13 constructs a diagram (e.g., a sphere) of the nodes that make up the NN from each node information contained in the NN information, and constructs a diagram (e.g., a line) of the edges that make up the NN from the edge information. The processing unit 13 also arranges the node diagram (e.g., a sphere) at the position in virtual space indicated by the node position information contained in each node information, arranges a diagram of an edge (e.g., a line) whose end point is the position in virtual space indicated by the edge position information contained in each edge information, and constructs a diagram that clearly shows that the node to which the edge is connected is connected to the diagram of the edge (e.g., a line).

Here, output is a concept that includes displaying on a display, projecting using a projector, printing on a printer, outputting sound, sending to an external device, storing on a recording medium, and passing on the processing results to other processing devices or other programs, etc.

The storage unit 11, the starting point storage unit 111, the goal storage unit 112, the reference score storage unit 113, and the NN storage unit 114 are preferably non-volatile recording media, but can also be realized with volatile recording media.

The process by which information is stored in the storage unit 11, etc. is not important. For example, information may be stored in the storage unit 11, etc. via a recording medium, information transmitted via a communication line, etc. may be stored in the storage unit 11, etc., or information inputted via an input device may be stored in the storage unit 11, etc.

The reception unit 12 and the information reception unit 121 are realized, for example, by wireless or wired communication means, means for receiving broadcasts, device drivers for input means such as a touch panel or keyboard, control software for a menu screen, etc.

The processing unit 13, image feature acquisition unit 131, sound feature acquisition unit 132, image score acquisition unit 133, sound score acquisition unit 134, total score acquisition unit 135, firing node determination unit 136, growth unit 138, state determination unit 137, and reference score change unit 139 can usually be realized by a processor, memory, etc. The processing procedures of the processing unit 13, etc. are usually realized by software, and the software is recorded in a recording medium such as a ROM. However, they may also be realized by hardware (dedicated circuitry). The processor may be a CPU, MPU, GPU, etc., and the type does not matter.

The output unit 14 can be realized, for example, by driver software for an output device, or by a combination of driver software for an output device and an output device, or by wireless or wired communication means or broadcasting means. The output device can be, for example, a display or a speaker.

Next, an example of the operation of the NN growth device 1 will be explained using the flowchart in Figure 2.

(Step S201) The information receiving unit 121 determines whether or not information has been received. If information has been received, the process proceeds to step S202; if information has not been received, the process returns to step S201.

(Step S202) The processing unit 13 determines whether image information is present in the information received in step S201. If image information is present, the process proceeds to step S203; if not, the process proceeds to step S204.

(Step S203) The image feature acquisition unit 131 acquires one or more pieces of image feature information from the image information accepted in step S201, and temporarily stores the information in a buffer (not shown).

(Step S204) The processing unit 13 determines whether or not sound information is present in the information received in step S201. If sound information is present, the process proceeds to step S205; if not, the process proceeds to step S206.

(Step S205) The sound feature acquisition unit 132 acquires one or more pieces of sound feature information from the sound information accepted in step S201.

(Step S206) The total score acquisition unit 135 etc. acquires the total score. An example of such score acquisition processing will be explained using the flowchart in FIG. 3.

(Step S207) The growth unit 138 etc. performs growth processing. An example of the growth processing is explained using the flowchart in FIG. 4.

(Step S208) The standard score change unit 139 performs processing to change the standard score. Return to step S201. An example of such standard change processing will be described using the flowchart in FIG. 16.

In the flowchart in Figure 2, processing ends when the power is turned off or an interrupt occurs to end processing.

Next, an example of the score acquisition process in step S206 will be explained using the flowchart in Figure 3.

(Step S301) The image score acquisition unit 133 determines whether image feature information is present in the acquired feature information. If image feature information is present, the process proceeds to step S302; if not, the process proceeds to step S304.

(Step S302) The image score acquisition unit 133 acquires one or more pieces of image feature information from the acquired feature information from a buffer (not shown).

(Step S303) The image score acquisition unit 133 acquires an image score using one or more pieces of image feature information acquired in step S302.

(Step S304) The sound score acquisition unit 134 determines whether or not sound feature information is present in the acquired feature information. If sound feature information is present, the process proceeds to step S305; if not, the process proceeds to step S307.

(Step S305) The sound score acquisition unit 134 acquires one or more pieces of sound feature information from the acquired feature information from a buffer (not shown).

(Step S306) The sound score acquisition unit 134 acquires a sound score using one or more pieces of sound feature information acquired in step S302.

(Step S307) The total score acquisition unit 135 acquires a total score using one or more of the acquired image scores or sound scores. It then returns to the upper level process.

The one or more pieces of feature information acquired are one or more types of information selected from the group consisting of one or more pieces of image feature information acquired in step S302 and one or more pieces of sound feature information acquired in step S305.

In the flowchart of FIG. 3, the overall score acquisition unit 135 may acquire the overall score using one or more pieces of acquired feature information, without using the image score and sound score.

Next, an example of the growth process in step S207 will be explained using the flowchart in Figure 4.

(Step S401) The firing node determination unit 136 performs a process to determine the node that will be fired initially. An example of such an initial firing node determination process will be described using the flowchart in FIG. 5.

(Step S402) The firing node determination unit 136 assigns 1 to counter i.

(Step S403) The ignition node determination unit 136 determines whether or not the i-th initial ignition node exists among the initial ignition nodes determined in step S401. If the i-th initial ignition node exists, the process proceeds to step S404; if not, the process proceeds to step S406.

(Step S404) The ignition node determination unit 136 performs ignition transmission processing. An example of the ignition transmission processing is described using the flowchart in FIG. 6.

(Step S405) The ignition node determination unit 136 increments the counter i by 1. Return to step S403.

(Step S406) The state determination unit 137 acquires the difference information. An example of the difference information acquisition process is described using the flowchart in FIG. 8.

(Step S407) The state determination unit 137 acquires a state identifier corresponding to the difference information acquired in step S406. Note that the state determination unit 137, for example, refers to a correspondence table having two or more pieces of correspondence information indicating the correspondence between the difference information conditions and the state identifiers, and acquires a state identifier that pairs with the condition that matches the acquired difference information. In addition, the correspondence table is stored in the storage unit 11.

(Step S408) The growth unit 138 assigns 1 to counter j.

(Step S409) The growth unit 138 determines whether or not the jth ignition node exists among the nodes that were ignited in steps S401 to S405. If the jth ignition node exists, the process proceeds to step S408; if not, the process returns to the upper level process.

(Step S410) The growth unit 138 increments the number of times information that is paired with the node identifier of the j-th firing node by 1.

(Step S411) The growth unit 138 performs single growth processing on the j-th firing node. An example of single growth processing is explained using the flowchart in FIG. 9.

Note that a single growth process is a growth process for one firing node and the edge connected to that node.

(Step S412) The growth unit 138 increments the counter j by 1. Return to step S407.

Next, an example of the initial firing node determination process in step S401 will be explained using the flowchart in Figure 5.

(Step S501) The firing node determination unit 136 assigns 1 to counter i.

(Step S502) The ignition node determination unit 136 determines whether or not the i-th initial ignition condition exists in the start point storage unit 111. If the i-th initial ignition condition exists, the process proceeds to step S503; if it does not exist, the process returns to the upper level process.

(Step S503) The firing node determination unit 136 obtains the i-th initial firing condition from the starting point storage unit 111.

(Step S504) The ignition node determination unit 136 acquires one or more pieces of feature information to be used in determining the i-th initial ignition condition from a buffer (not shown). Each of the one or more pieces of feature information is, for example, image feature information or sound feature information.

(Step S505) The ignition node determination unit 136 judges whether or not the one or more pieces of feature information acquired in step S504 satisfy the i-th initial ignition condition acquired in step S503. If the i-th initial ignition condition is satisfied, the process proceeds to step S506; if not, the process proceeds to step S509.

(Step S506) The ignition node determination unit 136 refers to the NN storage unit 114 and determines whether or not a ignition probability exists for the ignition node identifier that is paired with the i-th initial ignition condition. If a ignition probability exists, the process proceeds to step S507; if not, the process proceeds to step S508.

(Step S507) The ignition node determination unit 136 uses the ignition probability of the i-th initial ignition condition and the ignition node identifier that is paired with it to determine whether or not it will ignite this time. If it will ignite, the process proceeds to step S508, and if it will not ignite, the process proceeds to step S509.

(Step S508) The ignition node determination unit 136 acquires ignition information having an ignition node identifier that pairs with the i-th initial ignition condition, and stores the ignition information in the storage unit 11.

(Step S509) The ignition node determination unit 136 increments the counter i by 1. Return to step S502.

Next, an example of the ignition transmission process in step S404 will be described using the flowchart in FIG. 6.

(Step S601) The firing node determination unit 136 acquires all edge information including the firing node identifier as the identifier of the connection source node from the NN storage unit 114. Note that the firing node identifier here is, for example, the node identifier of the i-th initial firing node in step S403.

(Step S602) The firing node determination unit 136 assigns 1 to counter i.

(Step S603) The ignition node determination unit 136 determines whether the i-th edge information exists among the edge information acquired in step S601. If the i-th edge information exists, the process proceeds to step S604; if not, the process returns to the upper level process.

(Step S604) The firing node determination unit 136 judges whether or not the node identifier of another node is present in the i-th edge information. If the node identifier of another node is present, the process proceeds to step S605; if not, the process proceeds to step S612. Note that the node identifier of another node in the edge information is the node identifier of the node to which the edge is connected.

(Step S605) The firing node determination unit 136 acquires the node identifier of another node in the i-th edge information. Next, the firing node determination unit 136 acquires the node information of the node identified by the node identifier from the NN storage unit 114.

(Step S606) The ignition node determination unit 136 uses the node information acquired in step S605 to determine whether or not the node corresponding to the node information will ignite. An example of such ignition determination processing will be described with reference to the flowchart in FIG. 7.

(Step S607) If the determination result in step S606 is "fire", the ignition node determination unit 136 proceeds to step S608, and if the determination result is "do not ignite", the ignition node determination unit 136 proceeds to step S612.

(Step S608) The ignition node determination unit 136 acquires ignition information having the node identifier included in the node information acquired in step S605, and stores the ignition information in the storage unit 11.

(Step S609) The ignition node determination unit 136 changes the ignition probability information contained in the node information acquired in step S605. Here, the ignition node determination unit 136 changes the ignition probability information so that the ignition probability specified by the ignition probability information increases. Note that the amount of increase does not matter.

(Step S610) The firing node determination unit 136 judges whether or not to end the transmission of firing (which can also be called the transmission of information) between nodes. If the transmission is to be ended, the process proceeds to step S612, and if the transmission is not to be ended, the process proceeds to step S611. Note that the transmission is to be ended, for example, when the node in question is the terminal node in the NN.

(Step S611) The ignition node determination unit 136 performs ignition transmission processing with the node in question as the node of interest. An example of the ignition transmission processing is shown in FIG. 6.

(Step S612) The ignition node determination unit 136 increments the counter i by 1. Return to step S603.

In the flowchart of FIG. 6, when transmitting firing (transmission of information) between nodes, the firing node determination unit 136 preferably updates the retained energy amount information by subtracting the amount of energy indicated by the retained energy amount information held by the node information that caused the firing. This may also be applied to the retained energy amount information paired with the AXON identifier of the AXON used for the transmission, and the retained energy amount information paired with the Dendrites identifier of the Dendrites used for the transmission. The function for reducing the amount of energy is stored, for example, in the storage unit 11. The function in question is not important. As the function is a publicly known technology, a detailed description will be omitted.

In addition, in the flowchart of FIG. 6, the ignition node determination unit 136 typically performs processing to pass one or more pieces of feature information received by the ignition source node to the destination node where the ignition will occur.

Next, an example of the ignition determination process in step S606 will be explained using the flowchart in Figure 7.

(Step S701) The firing node determination unit 136 obtains the firing conditions corresponding to the node information obtained in step S605.

(Step S702) The ignition node determination unit 136 acquires one or more pieces of feature information. Note that the one or more pieces of feature information are passed from the node that is the source of the ignition.

(Step S703) The ignition node determination unit 136 judges whether or not one or more pieces of feature information acquired in step S702 satisfy the ignition condition acquired in step S701. If the ignition condition is satisfied, the process proceeds to step S704, and if the ignition condition is not satisfied, the process proceeds to step S707.

(Step S704) The firing node determination unit 136 judges whether the node information of interest has firing probability information. If it has firing probability information, the process proceeds to step S705, and if it does not have firing probability information, the process proceeds to step S706.

(Step S705) The ignition node determination unit 136 acquires ignition probability information contained in the node information of interest. Next, the ignition node determination unit 136 uses the ignition probability information to determine whether or not ignition will occur. If ignition will occur, the process proceeds to step S706, and if ignition will not occur, the process proceeds to step S707.

(Step S706) The ignition node determination unit 136 assigns "ignite" to the judgment result. Returns to the upper level process.

(Step S707) The firing node determination unit 136 assigns "do not fire" to the judgment result. It returns to the upper level process.

Next, an example of the difference information acquisition process in step S406 will be explained using the flowchart in Figure 8.

(Step S801) The state determination unit 137 obtains the total score obtained in step S206.

(Step S802) The state determination unit 137 assigns 1 to counter i.

(Step S803) The state determination unit 137 judges whether the i-th firing pattern exists in the standard score storage unit 113. If the i-th firing pattern exists, the process proceeds to step S804. If it does not exist, the process proceeds to step S809.

(Step S804) The state determination unit 137 acquires the i-th firing pattern.

(Step S805) The state determination unit 137 refers to one or more pieces of ignition information in the storage unit 11 and determines whether the ignition pattern acquired in step S804 is satisfied. If the ignition pattern is satisfied, the process proceeds to step S806; if not, the process proceeds to step S808.

(Step S806) The state determination unit 137 obtains the reference score that is paired with the i-th firing pattern from the reference score storage unit 113.

(Step S807) The state determination unit 137 obtains difference information using the obtained total score and the obtained reference score. It then returns to the upper level process.

(Step S808) The state determination unit 137 increments the counter i by 1. Return to step S803.

(Step S809) The state determination unit 137 obtains the default standard score from the standard score storage unit 113. Go to step S807.

Next, single growth processing is performed in step S509. An example of single growth processing is explained using the flowchart in Figure 9.

(Step S901) The growth unit 138 performs edge growth processing. An example of the edge growth processing is explained using the flowchart in FIG. 10.

(Step S902) The growth unit 138 performs edge generation processing. An example of the edge generation processing is described using the flowchart in FIG. 12.

(Step S903) The growth unit 138 performs node generation processing. Then, the process returns to the upper level. An example of the node generation processing is explained using the flowchart in FIG. 14.

This may also be done in the flowcharts in Figures 1 to 5.

Next, an example of the edge growth process in step S901 will be explained using the flowchart in Figure 10.

(Step S1001) The growth unit 138 obtains node information identified by the node identifier of interest from the NN storage unit 114.

(Step S1002) The growth unit 138 assigns 1 to counter i.

(Step S1003) The growth unit 138 determines whether or not the i-th edge information exists in the NN storage unit 114 among the edge information of the edge whose origin is the node identified by the node identifier of interest. If the i-th edge information exists, the process proceeds to step S1004, and if not, the process returns to the upper level process.

(Step S1004) The growth unit 138 obtains the i-th edge information from the NN storage unit 114.

(Step S1005) The growth unit 138 determines whether or not a node is connected to the end of the edge corresponding to the i-th edge information. More specifically, the growth unit 138 determines whether or not the node identifier contained in the i-th edge information is only the focus node identifier. If it is only the focus node identifier, the process proceeds to step S1006; if it is not only the focus node identifier (if there are two node identifiers), the process proceeds to step S1010. Note that edge information containing only the focus node identifier is connected to the focus node and is information about an edge that can grow. On the other hand, edge information containing two node identifiers is information about an edge that cannot grow.

(Step S1006) The growth unit 138 acquires the second edge growth conditions.

(Step S1007) The growth unit 138 acquires the difference information acquired in step S207.

(Step S1008) The growth unit 138 determines whether the difference information acquired in step S1007 satisfies the second edge growth condition. If the second edge growth condition is satisfied, the process proceeds to step S1009; if not, the process proceeds to step S1014.

(Step S1009) The growth unit 138 performs edge extension processing. Proceed to step S1014. An example of edge extension processing will be explained using the flowchart in FIG. 10.

Note that edge extension processing is processing that extends the length of an edge, and is usually processing that changes edge position information or processing that adds the node identifier of the connected node to edge information.

(Step S1010) The growth unit 138 acquires the first edge growth conditions.

(Step S1011) The growth unit 138 acquires the difference information acquired in step S207.

(Step S1012) The growth unit 138 determines whether the difference information acquired in step S1011 satisfies the first edge growth condition. If the first edge growth condition is satisfied, the process proceeds to step S1013; if not, the process proceeds to step S1014.

(Step S1013) The growth unit 138 changes the weight of the i-th edge information to an increased weight.

(Step S1014) The growth unit 138 increments the counter i by 1. Return to step S1003.

In the flowchart of FIG. 10, the edge growth process may be replaced with a dendrites growth process for the dendrites that make up the edge, or an AXON growth process for the AXON.

Next, an example of the edge extension process in step S1009 will be explained using the flowchart in Figure 11.

(Step S1101) The growth unit 138 acquires edge position information contained in the acquired edge information.

(Step S1102) The growth unit 138 acquires goal information.

(Step S1103) The growth unit 138 uses the edge position information acquired in step S1101 and the goal information acquired in step S1102 to acquire new edge position information and update the edge position information. The growth unit 138 acquires position information from the edge position information that identifies the position in the direction indicated by the goal information.

The growth unit 138 acquires edge position information indicating a position that is a predetermined distance away from the position indicated by the edge position information acquired in step S1101 in the direction specified by the goal information, for example. When another node exists in the direction specified by the goal information from the position indicated by the edge position information acquired in step S1101, for example, the growth unit 138 acquires edge position information indicating a position that is between the position indicated by the edge position information and the node position information of the other node in the direction specified by the goal information, and that is a distance within a predetermined distance. When another node exists in the direction specified by the goal information from the position indicated by the edge position information acquired in step S1101, for example, the growth unit 138 acquires the node position information of the other node as edge position information. In other words, when the growth unit 138 acquires new edge position information, it is sufficient to acquire edge position information in the direction specified by the goal information from the current edge position information, and the edge position information does not matter. Note that when the node position information of another node is acquired as edge position information, this is the case when the edge is connected to another node by edge extension processing, as described later. In such a case, the growth unit 138 may obtain the node identifier of the other node.

(Step S1104) The growth unit 138 assigns 1 to counter i.

(Step S1105) The growth unit 138 determines whether the i-th node information exists in the NN storage unit 114. If the i-th node information exists, the process proceeds to step S1106; if not, the process returns to the upper level process.

(Step S1106) The growth unit 138 obtains the node position information contained in the i-th node information.

(Step S1107) The growth unit 138 judges whether or not the node position information acquired in step S1106 satisfies the connection condition. If the connection condition is satisfied, the process proceeds to step S1108; if not, the process proceeds to step S1111. Note that the connection condition is a condition for an edge to be connected to a node. The connection condition is, for example, that the distance between the position indicated by the node position information acquired in step S1106 and the position indicated by the new edge position information acquired in step S1103 is within a threshold or is smaller than the threshold.

(Step S1108) The growth unit 138 obtains the node identifier contained in the i-th node information.

(Step S1109) The growth unit 138 changes the edge position information updated in step S1103 to the node position information contained in the i-th node information.

(Step S1110) The growth unit 138 adds the node identifier obtained in step S1108 to the acquired edge information. It then returns to the upper level processing.

(Step S1111) The growth unit 138 increments the counter i by 1. Return to step S1105.

In the flowchart of FIG. 11, the edge extension process may be replaced with dendrites extension process of the dendrites that make up the edge, or with AXON extension process of AXON.

Dendrites extension processing is processing that extends dendrites, and in the processing explanation using FIG. 11, edge information is replaced with dendrites information. AXON extension processing is processing that extends AXON, and in the processing explanation using FIG. 11, edge information is replaced with AXON information.

Also, in the flowchart of FIG. 11, after processing in step S1110, the process may proceed to step S1111. In such a case, one edge may branch and be connected to two or more nodes.

In addition, in step S1107 of the flowchart in FIG. 11, the distance between the new edge position information acquired in step S1103 and the node position information of each node may be calculated, and it may be determined whether the node position information of the node with the smallest distance satisfies the connection condition.

Next, an example of the edge generation process in step S902 will be described using the flowchart in FIG. 12.

(Step S1201) The growth unit 138 obtains node information identified by the node identifier of interest from the NN storage unit 114.

(Step S1202) The growth unit 138 acquires the edge generation conditions.

(Step S1203) The growth unit 138 reads the difference information that has already been acquired.

(Step S1204) The growth unit 138 determines whether the difference information acquired in step S1203 satisfies the edge generation condition. If the edge generation condition is satisfied, the process proceeds to step S1204; if not, the process returns to the upper level process.

(Step S1205) The growth unit 138 performs edge information generation processing. An example of the edge information generation processing is explained using the flowchart in FIG. 13.

(Step S1206) The growth unit 138 accumulates the edge information constructed in step S1205 in the NN storage unit 114. It then returns to the upper level processing.

Next, an example of the edge information generation process in step S1205 will be explained using the flowchart in Figure 13.

(Step S1301) The growth unit 138 obtains goal information corresponding to the state identifier obtained in step S208 from the goal storage unit 112.

(Step S1302) The growing unit 138 obtains edge position information of a new edge using the node position information of the node of interest (the node where the edge is to be generated) and the goal information obtained in step S1301. The growing unit 138 obtains edge position information of an edge that starts from the node position information and extends in the direction of the goal information. Note that, for example, the distance between the position indicated by the edge position information and the position indicated by the node position information may or may not be determined in advance.

The growth unit 138 acquires edge position information indicating a position a predetermined distance away from the position indicated by the node position information of the node of interest in the direction specified by the goal information, for example. When another node exists in the direction specified by the goal information from the position indicated by the node position information of the node of interest, for example, the growth unit 138 acquires the node position information of the other node as edge position information. In other words, here, the generated edge becomes an edge connecting the node corresponding to the node position information of the node of interest and the other node. When the distance between the position indicated by the node position information of the node of interest and the position indicated by the node position information of the other node is equal to or greater than a threshold value, for example, when another node exists in the direction specified by the goal information from the position indicated by the node position information of the node of interest, the growth unit 138 acquires node position information indicating a position a predetermined distance away from the position indicated by the node position information of the node of interest, and when the distance between the position indicated by the node position information of the node of interest and the position indicated by the node position information of the other node is equal to or less than a threshold value that is smaller than the threshold value, the growth unit 138 acquires the node position information of the other node as edge position information. In other words, the growth unit 138 only needs to obtain edge position information for an edge that starts from the node position information and extends in the direction of the goal information, and the edge position information is not important.

(Step S1303) The growing unit 138 obtains an edge identifier for the new edge. For example, the growing unit 138 generates a new edge identifier. For example, the growing unit 138 obtains an unused edge identifier from the set of edge identifiers.

(Step S1304) The growing unit 138 acquires the node identifier (target node identifier) of the node to which the edge is connected. Here, the growing unit 138 acquires the node identifier that pairs with the node position information of the target node. The growing unit 138 may also acquire the node identifier of the newly connected node.

(Step S1305) The growth unit 138 constructs edge information having the edge identifier acquired in step S1303, the edge position information acquired in step S1302, and one or two node identifiers acquired in step S1304. It then returns to the upper level processing.

In the flowchart of FIG. 13, the edge corresponding to the generated edge information may be in a situation where there is no node connected ahead, or there may be a node connected ahead.

If there is a node connected to the edge corresponding to the generated edge information, the growing unit 138 acquires the node identifier that pairs with the node position information of the position in the direction of the goal information as the node identifier of the connected node. Note that when the growing unit 138 always configures and stores edge information such that a generated edge always connects two nodes, edge growing processing is not normally performed.

Next, an example of the node generation process in step S903 will be explained using the flowchart in Figure 14.

(Step S1401) The growth unit 138 acquires difference information.

(Step S1402) The growth unit 138 acquires the node generation conditions.

(Step S1403) The growth unit 138 determines whether the difference information acquired in step S1401 satisfies the node generation condition acquired in step S1402. If the node generation condition is satisfied, the process proceeds to step S1404, and if the node generation condition is not satisfied, the process returns to the upper level process.

(Step S1404) The growth unit 138 performs node information generation processing. An example of the node information generation processing is explained using the flowchart in FIG. 15.

(Step S1405) The growth unit 138 accumulates the node information constructed in step S1404 in the NN storage unit 114. It then returns to the upper level processing.

Next, an example of the node information generation process in step S1404 will be explained using the flowchart in Figure 15.

(Step S1501) The growth unit 138 acquires node position information contained in the focus node information identified by the focus node identifier.

(Step S1502) The growth unit 138 obtains goal information corresponding to the state identifier obtained in step S407 from the goal storage unit 112.

(Step S1503) The growth unit 138 uses the node position information acquired in step S1501 and the goal information acquired in step S1502 to acquire node position information indicating a position in the direction specified by the goal information relative to the position indicated by the node position information acquired in step S1501. This node position information is the position information of the new node.

The growth unit 138 acquires, for example, node position information indicating a position that is a predetermined distance away from the position indicated by the node position information acquired in step S1501 in the direction specified by the goal information. For example, if there is another node in the direction specified by the goal information from the position indicated by the node position information acquired in step S1501, the growth unit 138 acquires node position information indicating a position that is between the position indicated by the node position information acquired in step S1501 and the node position information of the other node in the direction specified by the goal information and that is within a predetermined distance. In other words, when the growth unit 138 acquires node position information of a new node, it is sufficient to acquire node position information in the direction specified by the goal information, and the node position information is not important.

(Step S1504) The growth unit 138 obtains a node identifier for the new node. The growth unit 138 generates a new node identifier. However, the growth unit 138 may obtain an unused node identifier from the set of node identifiers.

(Step S1505) The growth unit 138 acquires information contained in the node information of the node of interest, which is used for the node information of the new node. Such information includes, for example, ignition conditions, ignition probability information, and possessed energy amount information.

(Step S1506) The growth unit 138 constructs node information having the node identifier acquired in step S1504, the node position information acquired in step S1503, and the information acquired in step S1505. It then returns to the upper-level processing.

Next, an example of the criteria change process in step S208 will be explained using the flowchart in Figure 16.

(Step S1601) The standard score change unit 139 obtains the score change conditions from the storage unit 11.

(Step S1602) The reference score change unit 139 obtains the total score or difference information from a buffer (not shown).

(Step S1603) The reference score change unit 139 determines whether the total score or difference information satisfies the score change condition. If the score change condition is met, the process proceeds to step S1604; if not, the process returns to the upper level process.

(Step S1604) The standard score change unit 139 determines whether or not the i-th firing pattern exists in the standard score storage unit 113. If the i-th firing pattern exists, the process proceeds to step S1605; if not, the process returns to the upper level process.

(Step S1605) The reference score modification unit 139 assigns 1 to the counter i.

(Step S1606) The standard score change unit 139 obtains the i-th firing pattern from the standard score storage unit 113.

(Step S1607) The standard score change unit 139 refers to the firing information (information on the fired node) in the storage unit 11 and determines whether the i-th firing pattern is satisfied. If the i-th firing pattern is satisfied, the process proceeds to step S1608; if not, the process proceeds to step S1610. Note that if the i-th firing pattern is satisfied, this is usually the case when all node identifiers included in the firing pattern are included in the set of node identifiers of the fired node. However, if the i-th firing pattern is satisfied, it may also be the case, for example, that a threshold or higher percentage of node identifiers included in the firing pattern are included in the set of node identifiers of the fired node.

(Step S1608) The standard score modification unit 139 obtains the standard score that is paired with the i-th firing pattern from the standard score storage unit 113. The standard score modification unit 139 uses the obtained standard score to obtain the modified standard score.

If the status identifier is "positive", the revised standard score will usually be greater than the original standard score. If the status identifier is "negative", the revised standard score will usually be less than the original standard score.

(Step S1609) The standard score change unit 139 changes the standard score paired with the i-th firing pattern to the standard score obtained in step S1608.

(Step S1610) The reference score modification unit 139 increments the counter i by 1. Return to step S1606.

In the flowchart of FIG. 16, the score change condition may be different for each firing pattern. In such a case, if the score change condition paired with the firing pattern determined to be satisfied in step S1607 is satisfied, the score change condition paired with the firing pattern is changed.

In addition, in the flowchart of FIG. 16, the standard score change unit 139 may change the default standard score.

As described above, according to this embodiment, it is possible to simulate brain growth. In particular, according to the NN growth device 1, it is possible to simulate, for example, the growth of a person's brain from infancy onwards.

The processing in this embodiment may be realized by software. This software may be distributed by software download or the like. This software may also be recorded on a recording medium such as a CD-ROM and distributed. This also applies to the other embodiments in this specification. The software that realizes the NN growth device 1 in this embodiment is a program such as the following. That is, this program includes a computer that can access an NN storage unit in which neural network information having two or more pieces of node information each having a node identifier and one or more pieces of edge information each having an edge identifier and identifying a connection between the nodes is stored, a start point storage unit in which one or more pieces of ignition start point information each having one or more node identifiers that identify a node that ignites without passing through other nodes is stored in association with an initial ignition condition, which is a condition related to one or more types of feature information out of one or more pieces of image feature information for image information and one or more pieces of sound feature information for sound information, and which is a condition for determining a node that ignites without passing through other nodes, and a standard score storage unit that stores standard scores related to positivity and negativity, an information receiving unit that receives image information and sound information, an image feature acquisition unit that uses the image information received by the information receiving unit to acquire one or more pieces of image feature information for the image information, and a process for acquiring the image feature information using the sound information received by the information receiving unit. A program for causing the program to function as a growth unit that performs a growth process that grows edge information of one or more edges that are connected to nodes identified by the one or more node identifiers by edges and that are passed one or more types of information of the one or more image feature information and the one or more sound feature information, based on difference information regarding the difference between the total score and the reference score. A sound feature acquisition unit that acquires one or more sound feature information for the sound information; a total score acquisition unit that acquires a total score based on the one or more image feature information and the one or more sound feature information; an ignition node determination unit that determines one or more node identifiers that are paired with an initial ignition condition in which one or more types of information of the one or more image feature information and the one or more sound feature information match from the starting point storage unit, and determines the node identifier of a node that will ignite, the node being connected to the node identified by the one or more node identifiers by edges and that is passed one or more types of information of the one or more image feature information and the one or more sound feature information; and a growth unit that performs a growth process that grows edge information of one or more edges that are connected to nodes identified by one or more node identifiers of the one or more node identifiers determined by the ignition node determination unit, based on difference information regarding the difference between the total score and the reference score.

(Embodiment 2)
In this embodiment, an information processing device will be described that uses neural network information generated by a neural network growing device 1 to obtain firing patterns for received image information and/or sound information, and outputs information for the firing patterns.

In addition, in this embodiment, we will explain an information processing device in which the speed at which information is transmitted between nodes changes depending on temperature information.

FIG. 17 is a block diagram of the information processing device 2 in this embodiment. The information processing device 2 includes a storage unit 21, a reception unit 22, a processing unit 23, and an output unit 24. The storage unit 21 includes a starting point storage unit 111 and an NN storage unit 114. The reception unit 22 includes an information reception unit 221 and a temperature reception unit 222. The processing unit 23 includes a feature acquisition unit 231, an information transmission unit 232, an ignition pattern acquisition unit 233, and an output information acquisition unit 234. The output unit 24 includes an information output unit 241.

The storage unit 21 that constitutes the information processing device 2 stores various types of information. The various types of information include, for example, neural network information, one or more pieces of ignition start information, and one or more pieces of output management information.

The ignition start information is information that has an information identifier that identifies the characteristic information of the received information, and one or more node identifiers that identify the node that will be ignited first when the characteristic information is received.

The reception information is information received by the information reception unit 221. The reception information includes one or two types of information, image information or sound information. The reception information may be two or more types of information. The reception information may include, for example, tactile information and smell information. Tactile information is information relating to the sense of touch. Smell information is information relating to smell.

Output management information is information that has output conditions and output information. Output management information may be a pair of information of output conditions and output information.

An output condition is a condition used to determine output information. An output condition is a condition for output using a firing pattern. An output condition may be the firing pattern itself, or may be information having a firing pattern and output probability information. Output probability information is information regarding the probability of obtaining output information. An output condition may be information regarding a firing pattern and a lower limit of the number of node identifiers that the firing pattern to be applied has, or information regarding a lower limit of the ratio of node identifiers that the firing pattern to be applied has, etc. An firing pattern has one or more node identifiers. An firing pattern is a pattern of firing of one or more nodes. Output information is information corresponding to an firing pattern.

The output information is, for example, emotional information related to a person's emotions, behavioral information related to a person's bodily movements, etc. Emotional information is, for example, happy, sad, frightened, surprised, etc. Emotional information is, for example, an ID that identifies an emotion. Emotional information may also be the above-mentioned state identifier. Behavioral information is, for example, information reflected in the movements of an avatar (character). Behavioral information is, for example, information reflected in the movements of an infant avatar. Note that the technology for moving an avatar is publicly known technology, so a detailed explanation will be omitted.

The output condition may be a condition that uses the firing pattern and one or more pieces of information related to external information. External information is information from the outside. External information may also be called user context. Examples of external information include temperature, weather, smell, sound, light, etc.

The NN storage unit 114 stores the neural network information accumulated by the NN growth device 1.

The reception unit 22 receives various types of information. Examples of the various types of information include reception information and temperature information.

The information receiving unit 221 receives reception information. For example, the information receiving unit 221 acquires image information captured by a camera. The information receiving unit 221 may also accept sound information acquired by a microphone. The reception information may include both image information and sound information.

Here, reception is a concept that includes the reception of information acquired by devices such as cameras and microphones, the reception of information transmitted via wired or wireless communication lines, and the reception of information read from recording media such as optical disks, magnetic disks, and semiconductor memories.

The temperature reception unit 222 receives temperature information. Temperature information is information that specifies a temperature. The temperature is, for example, the temperature of the external environment.

Here, reception is a concept that includes the reception of information input from input devices such as a microphone, keyboard, mouse, or touch panel, the reception of information transmitted via a wired or wireless communication line, and the reception of information read from recording media such as optical disks, magnetic disks, and semiconductor memory.

The processing unit 23 performs various types of processing. For example, the various types of processing are performed by the feature acquisition unit 231, the information transmission unit 232, the firing pattern acquisition unit 233, and the output information acquisition unit 234.

The feature acquisition unit 231 uses the image information accepted by the information acceptance unit 221 to acquire one or more pieces of image feature information for the image information. The feature acquisition unit 231 uses the sound information accepted by the information acceptance unit 221 to acquire one or more pieces of sound feature information for the sound information.

The processing performed by the feature acquisition unit 231 may be the same as the processing performed by one or two of the components of the image feature acquisition unit 131 and the sound feature acquisition unit 132.

The information transmission unit 232 determines a node identifier corresponding to each of the one or more pieces of feature information acquired by the feature acquisition unit 231 from one or more pieces of ignition start point information. Such a node identifier is the identifier of the node that will ignite. Moreover, such a node identifier is the identifier of the node that will ignite in the first stage.

Next, the information transmission unit 232 determines the node identifier of the node that is connected by an edge to the node identified by the one or more determined node identifiers, is passed feature information, and is to be fired.

The information transmission unit 232 performs an information transmission process that passes characteristic information from one firing node to the next firing node. The next firing node is a node that is connected to the first firing node by an edge and is a node that is determined to fire.

The information transmission unit 232 acquires node information of nodes connected to one firing node by an edge. Next, the information transmission unit 232 judges whether or not the node information satisfies the firing condition. Then, the information transmission unit 232 composes and accumulates firing information that includes the node identifier of the node information that satisfies the firing condition.

For example, when the information transmission unit 232 determines that the node information satisfies the ignition condition, it determines whether or not the node information will ignite with the probability indicated by the ignition probability information of the node information. Then, when the information transmission unit 232 determines that the node information will ignite based on the probability indicated by the ignition probability information, for example, it configures and stores ignition information including the node identifier of the node information.

It is preferable that the information transmission unit 232 changes the processing time for performing the information transmission process according to the temperature information received by the temperature reception unit 222. For example, when the temperature information indicates a high temperature, the information transmission unit 232 performs the information transmission process faster than when the temperature information indicates a lower temperature. For example, when the temperature information indicates a low temperature, the information transmission unit 232 delays the information transmission process compared to when the temperature information indicates a higher temperature. For example, when a delay condition is met, the information transmission unit 232 delays the information transmission process for a predetermined time. Delaying the information transmission process means, for example, waiting for a predetermined time. The delay condition is a condition for delaying the process and is a condition based on the temperature information. The delay condition is, for example, "temperature information <= threshold" or "temperature information < threshold".

It is preferable for the information transmission unit 232 to increase the firing probability information contained in the node information of the firing node. This is because the more a node fires, the easier it becomes to fire that node.

The firing pattern acquisition unit 233 acquires a firing pattern using one or more node identifiers determined by the information transmission unit 232. A firing pattern is a collection of information that identifies one or more firing nodes. A firing pattern is usually information that identifies nodes that fire simultaneously. A firing pattern has one or more node identifiers.

It is preferable that the firing pattern acquisition unit 233 acquires a firing pattern having one or more node identifiers of the node that finally fired. Note that the node that finally fired is a node that, among the nodes that fired, did not transmit information to other nodes connected by edges.

The output information acquisition unit 234 acquires output information corresponding to the firing pattern acquired by the firing pattern acquisition unit 233. The output information acquisition unit 234, for example, refers to one or more pieces of output management information in the storage unit 21, and acquires output information corresponding to the output conditions satisfied by the firing pattern acquired by the firing pattern acquisition unit 233.

The output information acquisition unit 234, for example, refers to one or more pieces of output management information in the storage unit 21, and determines the firing pattern of the output condition that is satisfied by the firing pattern acquired by the firing pattern acquisition unit 233. Next, the output information acquisition unit 234 determines whether or not to acquire output information, for example, with a probability based on the output probability information that pairs with the determined firing pattern, and if it is determined to acquire the output information, acquires the output information contained in the output management information.

The output unit 24 outputs various information. The various information is, for example, output information.

The information output unit 241 outputs the output information acquired by the output information acquisition unit 234. Here, output is a concept that includes display on a display, projection using a projector, printing on a printer, sound output, transmission to an external device, storage on a recording medium, and delivery of processing results to other processing devices, other programs, etc.

The storage unit 21 and the NN storage unit 114 are preferably non-volatile recording media, but can also be realized using volatile recording media.

The process by which information is stored in the storage unit 21, etc. is not important. For example, information may be stored in the storage unit 21, etc. via a recording medium, information transmitted via a communication line, etc. may be stored in the storage unit 21, etc., or information inputted via an input device may be stored in the storage unit 21, etc.

The reception unit 22, the information reception unit 221, and the temperature reception unit 222 are realized, for example, by a camera, a microphone, a wireless or wired communication means, a means for receiving broadcasts, a device driver for an input means such as a touch panel or a keyboard, control software for a menu screen, etc.

The processing unit 23, feature acquisition unit 231, information transmission unit 232, firing pattern acquisition unit 233, and output information acquisition unit 234 can usually be realized by a processor, memory, etc. The processing procedures of the processing unit 23, etc. are usually realized by software, and the software is recorded in a recording medium such as a ROM. However, they may also be realized by hardware (dedicated circuitry). The processor may be a CPU, MPU, GPU, etc., and the type is not important.

The output unit 24 and the information output unit 241 may or may not include output devices such as a display or a speaker. The output unit 24, etc. may be realized by driver software for an output device, or by driver software for an output device and an output device, etc. The output unit 24, etc. may be configured by a robot, for example.

Next, an example of the operation of the information processing device 2 will be described using the flowchart in FIG. 18.

(Step S1801) The information reception unit 221 determines whether reception information has been received. If reception information has been received, the process proceeds to step S1802, and if reception information has not been received, the process returns to step S1801.

(Step S1802) The feature acquisition unit 231 acquires one or more pieces of feature information from the reception information accepted in step S1801. For example, the feature acquisition unit 231 acquires one or more pieces of feature information from the image information accepted in step S1801. For example, the feature acquisition unit 231 acquires one or more pieces of feature information from the sound information accepted in step S1801.

(Step S1803) The temperature reception unit 222 acquires the temperature information.

(Step S1804) The information transmission unit 232 performs information transmission processing within the neural network. An example of the information transmission processing is explained using the flowchart in FIG. 19.

(Step S1805) The firing pattern acquisition unit 233 acquires a firing pattern having one or more node identifiers of the node that fired last in step S1804. Note that the node that fired last is a node that fired and did not pass on feature information to other nodes.

(Step S1806) The output information acquisition unit 234 acquires output information corresponding to the firing pattern acquired in step S1805.

(Step S1807) If the output information was obtained in step S1806, proceed to step S1808; if not, return to step S1801.

(Step S1808) The information output unit 241 outputs the information acquired in step S1806. Return to step S1801.

In the flowchart of FIG. 18, the processing unit 23 may have a state determination unit 137 and a growth unit 138, and may perform the growth process described above.

In addition, in the flowchart in Figure 18, processing ends when the power is turned off or an interrupt occurs to end processing.

Next, an example of the information transmission process in step S1804 will be explained using the flowchart in Figure 19.

(Step S1901) The information transmission unit 232 assigns 1 to counter i.

(Step S1902) The information transmission unit 232 determines whether the i-th feature information exists among the feature information acquired in step S1802. If the i-th feature information exists, the process proceeds to step S1903; if it does not exist, the process returns to the upper level process.

(Step S1903) The information transmission unit 232 refers to one or more pieces of ignition start point information in the storage unit 21, and acquires a node identifier contained in each of the one or more pieces of ignition start point information that satisfy the i-th feature information. The information transmission unit 232 composes ignition information including the node identifier, and stores it in the storage unit 21. It is preferable that the information transmission unit 232 acquires timer information indicating the time of ignition from a clock (not shown), composes ignition information having the timer information and the node identifier, and stores it in the storage unit 21. The one or more node identifiers are identifiers of nodes that ignite in the first stage. Also, the node identifiers are ignition node identifiers. It is also possible that the information transmission unit 232 is not able to acquire ignition node identifiers.

(Step S1904) The information transmission unit 232 assigns 1 to counter j.

(Step S1905) The information transmission unit 232 determines whether or not the jth ignition node identifier exists among the ignition node identifiers acquired in step S1903. If the jth ignition node identifier exists, the process proceeds to step S1906; if not, the process proceeds to step S1910.

(Step S1906) The information transmission unit 232 performs a process of adding the i-th feature information to the node identified by the j-th firing node identifier. Note that the process of adding the i-th feature information to a node is, for example, a process of writing the i-th feature information to the node information of the node, and a process of associating the i-th feature information with the node information of the node.

(Step S1907) The information transmission unit 232 performs an update to increase the number of times information contained in the node information corresponding to the j-th ignition node identifier. For example, the information transmission unit 232 reads out the number of times information contained in the node information corresponding to the j-th ignition node identifier, and overwrites the number of times information by adding 1 to the read number of times information.

(Step S1908) The information transmission unit 232 performs the next transmission process using the j-th ignition node identifier as the node identifier of interest. An example of the next transmission process is described using the flowchart in FIG. 20.

The next transmission process is a process of transmitting characteristic information for the node identified by the target node identifier to the node that is connected by an edge to the node identified by the target node identifier and that fires. The process of transmitting characteristic information is an information transmission process.

(Step S1909) The information transmission unit 232 increments the counter j by 1. Return to step S1905.

(Step S1910) The information transmission unit 232 increments the counter i by 1. Return to step S1902.

Next, an example of the next transmission process in step S1908 will be explained using the flowchart in Figure 20.

(Step S2001) The information transmission unit 232 determines whether the acquired temperature information matches the delay condition. If the delay condition is met, the process proceeds to step S2002; if the delay condition is not met, the process proceeds to step S2003.

(Step S2002) The information transmission unit 232 WAITs. Note that it is preferable that the WAIT time be determined in advance, but this is not limiting.

(Step S2003) The information transmission unit 232 acquires all edge information including the ignition node identifier of interest from the NN storage unit 114.

(Step S2004) The information transmission unit 232 assigns 1 to counter i.

(Step S2005) The information transmission unit 232 determines whether the i-th edge information exists among the edge information acquired in step S2001. If the i-th edge information exists, the process proceeds to step S2006; if not, the process returns to the upper level process.

(Step S2006) The information transmission unit 232 determines whether or not the node identifier of another node is present in the i-th edge information. If the node identifier of another node is present, the process proceeds to step S2007; if not, the process proceeds to step S2014. Note that the node identifier of another node in the edge information is the node identifier of the node to which the edge is connected. If the node identifier of another node is present in the i-th edge information, this is the case when the edge is connected to two nodes.

(Step S2007) The information transmission unit 232 acquires the node identifier of another node in the i-th edge information. Next, the information transmission unit 232 acquires the node information of the node identified by the node identifier from the NN storage unit 114.

(Step S2008) The information transmission unit 232 uses the node information acquired in step S2007 to determine whether or not the node corresponding to the node information will ignite. An example of such ignition determination processing will be described with reference to the flowchart in FIG. 21.

(Step S2009) If the result of the determination in step S2008 is "fire", the information transmission unit 232 proceeds to step S2010, and if the result is "do not fire", the information transmission unit 232 proceeds to step S2014.

(Step S2010) The information transmission unit 232 acquires ignition information having the node identifier included in the node information acquired in step S2007, and stores the ignition information in the storage unit 11.

(Step S2011) The information transmission unit 232 changes the ignition probability information contained in the node information acquired in step S2007. Here, the information transmission unit 232 changes the ignition probability information so that the ignition probability specified by the ignition probability information increases.

(Step S2012) The information transmission unit 232 judges whether or not to end the transmission of information between nodes. If the transmission is to be ended, the process proceeds to step S2014, and if the transmission is not to be ended, the process proceeds to step S2013. Note that the transmission is to be ended, for example, when the node in question is the terminal node in the NN. Also, when the transmission is to be ended, the node identifiers contained in the firing information accumulated immediately before in step S2010 are the node identifiers that constitute the firing pattern.

(Step S2013) The information transmission unit 232 performs the next transmission process with the node in question as the node of interest. An example of the next transmission process is shown in FIG. 20.

(Step S2014) The information transmission unit 232 increments the counter i by 1. Return to step S2005.

In the flowchart of FIG. 20, when the information transmission unit 232 transmits information between nodes, it is preferable to update the retained energy amount information by subtracting the amount of energy indicated by the retained energy amount information held by the node information that caused the ignition. This may also be applied to the retained energy amount information paired with the AXON identifier of the AXON used for the transmission, and the retained energy amount information paired with the Dendrites identifier of the Dendrites used for the transmission. The function for reducing the amount of energy is stored, for example, in the storage unit 21. The function in question is not important. As the function is a publicly known technology, a detailed explanation will be omitted.

Next, an example of the ignition determination process in step S2008 will be explained using the flowchart in Figure 21.

(Step S2101) The information transmission unit 232 acquires the firing condition corresponding to the node information of step S2005. Note that the firing condition may be different for each node, or may be common to two or more nodes.

(Step S2102) The information transmission unit 232 acquires one or more pieces of feature information. Note that the one or more pieces of feature information here are pieces of feature information passed from the node that caused the ignition.

(Step S2103) The information transmission unit 232 determines whether or not the one or more pieces of feature information acquired in step S2102 satisfy the ignition condition acquired in step S2101. If the ignition condition is satisfied, the process proceeds to step S2104, and if the ignition condition is not satisfied, the process proceeds to step S2107.

(Step S2104) The information transmission unit 232 determines whether the node information of interest has firing probability information. If it has firing probability information, the process proceeds to step S2105, and if it does not have firing probability information, the process proceeds to step S2106.

(Step S2105) The information transmission unit 232 acquires the firing probability information contained in the node information of interest. Next, the information transmission unit 232 uses the firing probability information to determine whether or not firing will occur. If firing will occur, the process proceeds to step S2106, and if not, the process proceeds to step S2107.

(Step S2106) The information transmission unit 232 assigns "fire" to the judgment result. It returns to the upper level process.

(Step S2107) The information transmission unit 232 assigns "does not fire" to the judgment result. It returns to the upper level process.

As described above, according to this embodiment, it is possible to simulate the operation of a grown brain. In other words, the information processing device 2 can simulate the operation of the brain reacting to received information using the NN information constructed by the NN growth device 1. In particular, the information processing device 2 can simulate the operation of a person's brain from infancy onwards.

In this embodiment, the information processing device may realize the growth process performed by the NN growth device 1. In such a case, the information processing device can output output information for the received reception information while growing the neural network. In addition, the information processing device in such a case is the information processing device 3. In addition to the configuration of the information processing device 2, the processing unit 23 of the information processing device 3 has the image score acquisition unit 133, sound score acquisition unit 134, total score acquisition unit 135, ignition node determination unit 136, state determination unit 137, growth unit 138, and reference score change unit 139 of the NN growth device 1. A block diagram of the information processing device 3 in such a case is shown in FIG. 22.

In FIG. 22, the processing of the firing node determination unit 136 of the NN growth device 1 is performed by the information transmission unit 232. Also, the feature acquisition unit 231 is a combination of the image feature acquisition unit 131 and the sound feature acquisition unit 132.

The software that realizes the information processing device 2 in this embodiment is a program such as the following. In other words, this program is a program for making a computer that can access the NN storage unit in which the neural network information accumulated by the NN growing device 1 is stored function as an information receiving unit that receives received information, which is one or more pieces of information from image information or sound information; a feature acquisition unit that acquires one or more pieces of feature information for the received information received by the information receiving unit; an information transmission unit that determines the node identifier of a node that will ignite, which is a node identifier corresponding to each of the one or more pieces of feature information acquired by the feature acquisition unit, from a start point storage unit in which one or more pieces of firing start point information having an information identifier that identifies the feature information of the received information and one or more node identifiers that identify the node that will ignite first when the feature information is accepted, and that is connected by an edge to the node identified by each of the one or more node identifiers and is a node to which the feature information is passed, and that will ignite; an firing pattern acquisition unit that acquires a firing pattern using one or more node identifiers determined by the information transmission unit; an output information acquisition unit that acquires output information corresponding to the firing pattern acquired by the firing pattern acquisition unit; and an information output unit that outputs the output information.

FIG. 23 also shows the appearance of a computer that executes the programs described in this specification to realize the NN growth device 1, information processing device 2, and information processing device 3 of the various embodiments described above. The above-mentioned embodiments can be realized by computer hardware and a computer program executed thereon. FIG. 23 is an overview of this computer system 300, and FIG. 24 is a block diagram of system 300.

In FIG. 23, computer system 300 includes computer 301, which includes a CD-ROM drive, keyboard 302, mouse 303, monitor 304, microphone 305, and camera 306.

In FIG. 24, in addition to a CD-ROM drive 3012, computer 301 includes an MPU 3013, a bus 3014 connected to the CD-ROM drive 3012 etc., a ROM 3015 for storing programs such as a boot-up program, a RAM 3016 connected to the MPU 3013 for temporarily storing instructions for application programs and providing temporary storage space, and a hard disk 3017 for storing application programs, system programs, and data. Although not shown here, computer 301 may further include a network card that provides connection to a LAN.

A program that causes computer system 300 to execute functions such as the NN growth device 1 of the above-mentioned embodiment may be stored on CD-ROM 3101, inserted into CD-ROM drive 3012, and then transferred to hard disk 3017. Alternatively, the program may be sent to computer 301 via a network (not shown) and stored on hard disk 3017. The program is loaded into RAM 3016 when executed. The program may be loaded directly from CD-ROM 3101 or the network.

The program does not necessarily have to include an operating system (OS) or a third-party program that causes the computer 301 to execute functions such as the NN growth device 1 of the above-mentioned embodiment. The program only needs to include an instruction portion that calls appropriate functions (modules) in a controlled manner to obtain the desired results. How the computer system 300 operates is well known, and a detailed description will be omitted.

In addition, in the above program, the steps of transmitting information and receiving information do not include processing performed by hardware, such as processing performed by a modem or interface card in the transmission step (processing that can only be performed by hardware).

Furthermore, the computer that executes the above program may be a single computer or multiple computers. In other words, it may perform centralized processing or distributed processing.

Furthermore, in each of the above embodiments, it goes without saying that two or more communication means present in one device may be realized physically by one medium.

In addition, in each of the above embodiments, each process may be realized by centralized processing in a single device, or may be realized by distributed processing in multiple devices.

The present invention is not limited to the above-described embodiment, and various modifications are possible, and it goes without saying that these are also included within the scope of the present invention.

As described above, the NN growth device 1 of the present invention has the effect of simulating brain growth and is useful as an NN growth device, etc.

Claims

an NN storage unit in which neural network information is stored, the neural network information including two or more pieces of node information having node identifiers and one or more pieces of edge information having edge identifiers and specifying connections between the nodes;
a start point storage unit in which one or more pieces of ignition start point information having one or more node identifiers for identifying a node that is to be ignited without passing through other nodes are stored in association with an initial ignition condition, which is a condition related to one or more types of feature information among one or more pieces of image feature information for image information and one or more pieces of sound feature information for sound information, and which is a condition for determining a node that is to be ignited without passing through other nodes;
a standard score storage unit for storing standard scores regarding positive and negative;
an information receiving unit that receives image information and sound information;
an image feature acquisition unit that acquires one or more image feature information for the image information by using the image information accepted by the information acceptance unit;
a sound feature acquisition unit that acquires, using the sound information accepted by the information acceptance unit, one or more pieces of sound feature information for the sound information;
a comprehensive score acquisition unit that acquires a comprehensive score based on the one or more pieces of image feature information and the one or more pieces of sound feature information;
an ignition node determination unit that determines, from the starting point storage unit, one or more node identifiers that are paired with an initial ignition condition in which one or more types of information among the one or more image feature information and the one or more sound feature information match, and that determines a node identifier of a node to be ignited, the node being connected to the nodes identified by the one or more node identifiers by an edge and receiving one or more types of information among the one or more image feature information and the one or more sound feature information;
a growth unit that performs a growth process, which is a process of growing edge information of one or more edges that connect to nodes identified by one or more node identifiers among the one or more node identifiers determined by the ignition node determination unit, based on difference information regarding the difference between the total score and the reference score.
an image score acquisition unit that acquires an image score, which is a score regarding positive and negative, using the one or more pieces of image feature information;
a sound feature acquisition unit that acquires, using the sound information accepted by the information acceptance unit, one or more pieces of sound feature information for the sound information;
a sound score acquisition unit that acquires a sound score, which is a score regarding positive and negative, using the one or more pieces of sound feature information;
The overall score acquisition unit,
The apparatus of claim 1 wherein the image score and the sound score are used to obtain the overall score.
The edge information has weights,
The firing node determination unit
determining node identifiers of nodes connected by the edges and whose weights satisfy a weight-related propagation condition;
The growth portion is
The NN growth device of claim 1, further comprising: a first edge growth process that acquires difference information regarding the difference between the overall score and the reference score, and increases the weighting of the edge information of each of the one or more edges when the difference information matches a first edge growth condition.
The edge information may include edge position information indicating a position of an end of an edge;
The growth portion is
The NN growth device of claim 1, further comprising: acquiring difference information regarding the difference between the overall score and the reference score; and, when the difference information meets a second edge growth condition, performing a second edge growth process that changes the edge position information contained in the edge information of each of the one or more edges and extends the length of the edge.
a goal storage unit for storing goal information that identifies a goal corresponding to each of two or more states including positive and negative;
and a state determination unit that determines one state from the two or more states including positive and negative using the difference information acquired by the growth unit,
the goal information includes goal position information that identifies the position of the goal, or goal direction information that indicates the direction of the goal,
The growth portion is
The NN growth device of claim 4 performs the second edge growth process, which acquires the goal information that pairs with the one state determined by the state determination unit, changes the edge position information contained in the edge information of each of the one or more edges, acquires new edge position information in the direction indicated by the goal information, and accumulates the new edge position information.
The NN growth device according to claim 1, further comprising a standard score change unit that changes the standard score in the standard score storage unit based on the overall score acquired by the overall score acquisition unit.
The reference score changing unit,
7. The neural network growing apparatus according to claim 6, wherein the reference score in the reference score storage unit is changed only when the total score meets a score change condition.
The standard score storage unit includes:
storing a reference score paired with each of the one or more firing patterns using the one or more node identifiers;
The growth portion is
2. The NN growing device according to claim 1, further comprising: a standard score that is paired with an ignition pattern corresponding to one or more node identifiers determined by the ignition node determination unit; difference information regarding the difference between the total score and the standard score; and performing the growth process based on the difference information.
The firing node determination unit
2. The NN growing device of claim 1, which determines whether one or more pieces of feature information passed from one or more other nodes connected by edges satisfy a firing condition for one or more pieces of feature information, and determines a node identifier of a node determined to satisfy the firing condition.
The node information includes node position information that identifies a position of the node,
a goal storage unit for storing goal information that identifies a goal corresponding to each of two or more states including positive and negative;
and a state determination unit that determines one state from the two or more states including positive and negative using the difference information acquired by the growth unit,
The growth portion is
2. The NN growing device according to claim 1, wherein an edge generation process is performed to obtain difference information regarding a difference between the total score and the reference score, and based on the difference information, generate and accumulate edge information for edges extending from nodes identified by one or more node identifiers among the one or more node identifiers determined by the ignition node determination unit in a direction indicated by the goal information that pairs with the one state determined by the state determination unit.
The firing node determination unit
Storing frequency information relating to the frequency of the determined node identifier in association with the node identifier;
The growth portion is
11. The neural network growing apparatus according to claim 10, wherein the edge generation process is performed on a node identified by a node identifier corresponding to the number information that meets an edge generation condition.
the node is a soma,
The edge includes an AXON and a Dendrite.
The NN growing apparatus according to claim 1 , wherein the edge information includes AXON information having an AXON identifier and AXON position information indicating a position of the AXON, and Dendrites information having a Dendrites identifier and Dendrites position information indicating a position of the Dendrites.
a neural network storage unit for storing neural network information accumulated by the neural network growing device according to claim 1;
an information receiving unit that receives reception information which is one or more types of information of image information and sound information;
a characteristic acquisition unit that acquires one or more pieces of characteristic information for the received information received by the information receiving unit;
an information transmission unit which determines a node identifier of a node to be ignited, the node identifier being a node identifier corresponding to the one or more pieces of feature information acquired by the feature acquisition unit, from a start point storage unit in which one or more pieces of ignition start point information are stored, the start point information having an information identifier for identifying the feature information of the received information and one or more node identifiers for identifying a node that will be ignited without passing through other nodes when the feature information is accepted, and which is connected by an edge to the node identified by the one or more node identifiers and is a node to which the feature information is passed, the node identifier of the node to be ignited;
an ignition pattern acquisition unit that acquires an ignition pattern using one or more node identifiers determined by the information transmission unit;
an output information acquisition unit that acquires output information corresponding to the ignition pattern acquired by the ignition pattern acquisition unit;
and an information output unit that outputs the output information.
The temperature sensor further includes a temperature receiving unit that receives temperature information.
The information transmission unit is
The information processing device according to claim 13, further comprising: an information transmission process for transmitting the characteristic information corresponding to the fired node from the fired node to the next fired node; and a processing time for performing the information transmission process is changed depending on the temperature information received by the temperature receiving unit.
a starting point storage unit storing one or more ignition starting point information having one or more node identifiers for identifying nodes that ignite without passing through other nodes in association with an initial ignition condition, which is a condition related to one or more types of feature information among one or more image feature information for image information and one or more sound feature information for sound information, and which is a condition for determining a node that ignites without passing through other nodes; a standard score storage unit storing standard scores related to positive and negative; an information receiving unit; an image feature acquisition unit; a sound feature acquisition unit; a total score acquisition unit; an ignition node determination unit; and a growth unit,
an information receiving step in which the information receiving unit receives image information and sound information;
an image feature acquiring step in which the image feature acquiring unit acquires one or more image feature information for the image information by using the image information accepted in the information accepting step;
a sound feature acquiring step in which the sound feature acquiring unit acquires one or more pieces of sound feature information for the sound information by using the sound information accepted in the information accepting step;
a comprehensive score acquisition step in which the comprehensive score acquisition unit acquires a comprehensive score based on the one or more pieces of image feature information and the one or more pieces of sound feature information;
an ignition node determination step in which the ignition node determination unit determines, from the starting point storage unit, one or more node identifiers that are paired with an initial ignition condition in which one or more types of information among the one or more image feature information and the one or more sound feature information match, and determines a node identifier of a node to be ignited, the node being connected to the node identified by the one or more node identifiers by an edge and receiving one or more types of information among the one or more image feature information and the one or more sound feature information;
A method for producing neural network information comprising: a growing step in which the growing unit performs a growing process, which is a process of growing edge information of one or more edges that connect to nodes identified by one or more node identifiers among the one or more node identifiers determined in the ignition node determination step, based on difference information regarding the difference between the total score and the reference score.
a computer that can access an NN storage unit in which neural network information having two or more pieces of node information each having a node identifier and one or more pieces of edge information each having an edge identifier and specifying a connection between the nodes is stored; a start point storage unit in which one or more pieces of ignition start point information each having one or more node identifiers that identify a node that ignites without passing through other nodes is stored in association with an initial ignition condition, which is a condition related to one or more types of feature information among one or more pieces of image feature information for image information and one or more pieces of sound feature information for sound information, and which is a condition for determining a node that ignites without passing through other nodes; and a standard score storage unit that stores standard scores related to positive and negative,
an information receiving unit that receives image information and sound information;
an image feature acquisition unit that acquires one or more image feature information for the image information by using the image information accepted by the information acceptance unit;
a sound feature acquisition unit that acquires, using the sound information accepted by the information acceptance unit, one or more pieces of sound feature information for the sound information;
a comprehensive score acquisition unit that acquires a comprehensive score based on the one or more pieces of image feature information and the one or more pieces of sound feature information;
an ignition node determination unit that determines, from the starting point storage unit, one or more node identifiers that are paired with an initial ignition condition in which one or more types of information among the one or more image feature information and the one or more sound feature information match, and that determines a node identifier of a node to be ignited, the node being connected to the nodes identified by the one or more node identifiers by an edge and receiving one or more types of information among the one or more image feature information and the one or more sound feature information;
A program for functioning as a growth unit that performs a growth process, which is a process of growing edge information of one or more edges that are connected to nodes identified by one or more node identifiers among the one or more node identifiers determined by the ignition node determination unit, based on difference information regarding the difference between the overall score and the reference score.