Artificial intelligence feedback system and method

Abstract

The present invention provides an artificial intelligence feedback system, which includes a script extractor and executor for receiving an input 1 including basic materials for learning and an input 2 including core contents for learning and generating an initial condition for a popularity frequency value or a ranking value of information related to the input 2 from the input 1, an artificial intelligence connected to the script extractor and executor for generating a two-dimensional information table in a matrix form with a keyword and a core information according to the popularity frequency value or the ranking value derived from the initial condition as two axes, and adding the popularity frequency value or the ranking value to a corresponding coordinate of the two-dimensional information table to match knowledge acquired by the artificial intelligence to the corresponding coordinate, an output unit connected to the artificial intelligence and outputting a deep truth value derived by the artificial intelligence from the two-dimensional information table, and a repository connected to the artificial intelligence for organizing and storing the context of knowledge formed by linking the deep truth values ​​derived by the artificial intelligence by category.

Description

[Technical field]

[0001] The present invention relates to an artificial intelligence feedback system and method. [Background technology]

[0002] The learning method of general artificial intelligence requires many steps and a long time to accumulate patterns and realize them. The learning method of artificial intelligence requires that materials be unilaterally injected from the outside in order for the AI ​​to gain knowledge and recognize things, thereby finding a new equilibrium point. Conventional AI learning methods rely on a huge amount of input information and regression analysis classification based on that information, and therefore have the problem of requiring exponentially more calculations and longer times as the learning stage progresses. Summary of the Invention [Problem to be solved by the invention]

[0003] An object of the present invention is to provide an artificial intelligence feedback method and an artificial intelligence feedback system that efficiently provide intuitive feedback for learning.

[0004] The object of the present invention is not limited to the above-mentioned objects, and other objects not mentioned herein will be clearly understood by those skilled in the art from the following description. [Means for solving the problem]

[0005] In order to achieve the above object, an artificial intelligence feedback system according to one aspect of the present invention comprises: An artificial intelligence feedback system including a script extractor and executor, an artificial intelligence coupled to the script extractor and executor for acquiring knowledge, an output unit coupled to the artificial intelligence, and a repository coupled to the artificial intelligence, wherein the script extractor and executor is Input 1 contains the basic material to learn 、 and The artificial intelligence Input 2 including core contents for learning is provided, and the input 2 is related to the input 2. Characteristic Information It is a cumulative value Popularity Frequency Value 、 or Ranking information based on the distribution characteristics of the cumulative values.The initial condition for the rank value is From Input 1 Generate death , The artificial intelligence is The popularity frequency value or the ranking value derived from the initial conditions Using A two-dimensional information table in the form of a matrix is ​​generated with a keyword and core information as axes, and the popularity frequency value or the ranking value is added to the corresponding coordinates of the two-dimensional information table. The above Matching knowledge death , The output unit is The AI ​​outputs the deep truth value derived from the two-dimensional information table. death , The repository may include: The AI ​​connects the deep truth values ​​and organizes the knowledge context formed by the connection into categories and stores them. death , the artificial intelligence moreover, a pseudo-linear conversion table having the same area as the two-dimensional information table, in which the ranking value of the two-dimensional information table is digitized from the two-dimensional information table and each coordinate is randomly turned on or off; a pseudo-linear conversion for operating the two-dimensional pseudo-linear conversion table on the two-dimensional information table is performed a plurality of times to form a two-dimensional inherent characteristic table in which each coordinate has a characteristic vector; information of coordinates whose characteristic vectors are not changed among the coordinates of the two-dimensional inherent characteristic table is derived as the deep truth value; the two-dimensional inherent characteristic table is converted into one-dimensional linear channels of multiple generations to which the deep truth values ​​are mapped, and stored in the repository; and knowledge contexts formed by connecting the deep truth values ​​of each of the one-dimensional linear channels of multiple generations are organized by category and stored in the repository; In the two-dimensional information table The two-dimensional pseudo-linear conversion table When the pseudo-linear transformation is performed multiple times to calculate , Each time, the two-dimensional pseudo-linear conversion table direction of rotate Calculations , 180 degrees Invert Calculations and symmetry of the center point Calculations , symmetrical Calculations Any one of Performing arithmetic transformations .

[0006] Also, The artificial intelligence feedback system is The two-dimensional pseudo-linear conversion table When the pseudo-linear transformation is performed multiple times to calculate , The coordinate values ​​of each of the coordinates are On or off state case , after changing the on state to 1 and the off state to 0.1, The coordinates of the two-dimensional pseudo-linear conversion table are The above Either one Arithmetic Transformations Do .

[0007] Also, the input 1 may include a number of image files and the input 2 may include edge intensities.

[0008] Moreover, the input 1 may include a plurality of document files, and the input 2 may include set words.

[0009] In order to achieve the above object, an artificial intelligence feedback method according to another aspect of the present invention comprises: Artificial intelligence that learns knowledge Input 1 contains the basic material to learn 、 and The artificial intelligence Providing an input 2 containing core content for learning to a script extractor and executor, and said script extractor and executor generating a script associated with said input 2. Characteristic Information It is a cumulative value Popularity Frequency Value 、 or Ranking information based on the distribution characteristics of the cumulative values. The initial condition for the rank value is From Input 1 providing the artificial intelligence with the popularity frequency value or the ranking value derived by the artificial intelligence from the initial conditions; Using A step of generating a two-dimensional information table in a matrix form with keywords and core information as axes, respectively, and adding the popularity frequency value or the ranking value to the corresponding coordinates of the two-dimensional information table through interaction between the artificial intelligence and the script extraction and execution unit, and The abovea step of matching knowledge; a step of generating a two-dimensional pseudo-linear conversion table from the two-dimensional information table, in which the ranking value of the two-dimensional information table is quantified and each coordinate is randomly turned on or off, and which has the same area as the two-dimensional information table; a step of performing a pseudo-linear conversion on the two-dimensional information table by the artificial intelligence to operate the two-dimensional pseudo-linear conversion table a plurality of times to form a two-dimensional inherent characteristic table in which each coordinate has a characteristic vector, and deriving information of coordinates whose characteristic vectors are not changed among the coordinates of the two-dimensional inherent characteristic table as deep truth values; a step of converting the two-dimensional inherent characteristic table into one-dimensional linear channels of multiple generations to which the deep truth values ​​are mapped, storing the converted information in a repository, and organizing the knowledge contexts formed by connecting the deep truth values ​​of each of the one-dimensional linear channels of multiple generations into categories and storing the converted information in the repository, In the two-dimensional information table The two-dimensional pseudo-linear conversion table When the pseudo-linear transformation is performed multiple times to calculate , Each time, the two-dimensional pseudo-linear conversion table direction of rotate Calculations , 180 degrees Invert Calculations and symmetry of the center point Calculations , symmetrical Calculations Any one of Performing arithmetic transformations .

[0010] Also, The artificial intelligence feedback method includes: The two-dimensional pseudo-linear conversion table When the pseudo-linear transformation is performed multiple times to calculate , The coordinate values ​​of each of the coordinates are On or off state case , the on state is changed to 1, and the off state is changed to 0.1, The coordinates of the two-dimensional pseudo-linear conversion table are The above Either one Arithmetic Transformations Do . Effect of the Invention

[0011] The artificial intelligence feedback system and method of the present invention as described above provides one or more of the following advantages:

[0012] The present invention can provide an artificial intelligence feedback system that efficiently provides intuitive feedback for learning. [Brief description of the drawings]

[0013] [Figure 1] 2 is a flow chart illustrating an artificial intelligence feedback method according to an embodiment of the present invention.

[0014] [Diagram 2] FIG. 2 is an explanatory diagram of the artificial intelligence feedback method according to FIG. 1; [Diagram 3] FIG. 2 is an explanatory diagram of the artificial intelligence feedback method according to FIG. 1; [Figure 4] FIG. 2 is an explanatory diagram of the artificial intelligence feedback method according to FIG. 1; [Diagram 5] FIG. 2 is an explanatory diagram of the artificial intelligence feedback method according to FIG. 1; [Figure 6] FIG. 2 is an explanatory diagram of the artificial intelligence feedback method according to FIG. 1; [Figure 7] FIG. 2 is an explanatory diagram of the artificial intelligence feedback method according to FIG. 1;

[0015] [Figure 8] FIG. 2 illustrates an artificial intelligence feedback system according to another embodiment of the present invention.

[0016] [Figure 9] FIG. 1 is an illustration of an artificial intelligence feedback system according to another embodiment. [Figure 10] FIG. 1 is an illustration of an artificial intelligence feedback system according to another embodiment.

[0017] [Figure 11] 1 is an image illustrating an artificial intelligence feedback method and training by images using an artificial intelligence system.

[0018] [Figure 12]1 is an image illustrating an artificial intelligence feedback method and written training using an artificial intelligence system.

[0019] [Figure 13] FIG. 10 is an explanatory diagram of an artificial intelligence feedback system according to another embodiment based on FIG.

[0020] [Figure 14] FIG. 9 is an explanatory diagram of an artificial intelligence feedback system according to another embodiment based on FIG. 8 . [Figure 15] FIG. 9 is an explanatory diagram of an artificial intelligence feedback system according to another embodiment based on FIG. 8 . [Figure 16] FIG. 9 is an explanatory diagram of an artificial intelligence feedback system according to another embodiment based on FIG. 8 . [Figure 17] FIG. 9 is an explanatory diagram of an artificial intelligence feedback system according to another embodiment based on FIG. 8 . [Figure 18] FIG. 9 is an explanatory diagram of an artificial intelligence feedback system according to another embodiment based on FIG. 8 . [Figure 19] FIG. 9 is an explanatory diagram of an artificial intelligence feedback system according to another embodiment based on FIG. 8 .

[0021] [Figure 20] A figure showing the artificial intelligence training results reflecting the process shown in Figures 14 to 19. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The advantages and features of the present invention, as well as the methods of achieving them, will become more apparent with reference to the following detailed embodiments taken in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be embodied in various different forms, and the present embodiments are provided merely to complete the disclosure of the present invention and to fully inform those skilled in the art of the invention of the scope of the invention, and the present invention is defined only by the scope of the claims. Like reference numbers refer to like elements throughout the specification. Spatially relative terms such as "below," "beneath," "lower," "above," "upper," and the like, are used to facilitate describing the relationship of one object or element to another object or element as illustrated in the drawings.

[0023] Spatially relative terms should be understood as terms that include different orientations of objects relative to one another during use or operation in addition to the orientation shown in the drawings. For example, if the objects shown in the drawings were turned over, objects described as "below" or "beneath" other objects would then be placed "above" the other objects. Thus, the exemplary term "below" can encompass both an orientation of below and above. Objects may be oriented in other directions, so that spatially relative terms can be interpreted according to orientation.

[0024] Although first, second, etc. are used to describe various objects, components and / or sections, these objects, components and / or sections are not limited by these terms. These terms are used merely to distinguish one object, component or section from another object, component or section. Thus, a first object, a first component or a first section referred to hereinafter is a second object, a second component or a second section within the spirit of the present invention.

[0025] The terms used in the present specification are intended to describe the embodiments and are not intended to limit the present invention. In this specification, the singular forms include the plural forms unless the context clearly indicates otherwise. As used in the specification, "comprises" and / or "comprising" does not exclude the presence or addition of one or more other components, steps, operations and / or objects mentioned.

[0026] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the meaning commonly understood by one of ordinary skill in the art to which this invention belongs. Additionally, terms defined in commonly used dictionaries are not idealized or over-analyzed unless expressly and specifically defined.

[0027] Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings. In the description with reference to the accompanying drawings, the same reference numbers will be given to the same or corresponding components regardless of the drawing numbers, and duplicate descriptions related thereto will be omitted.

[0028] FIG. 1 is a flow chart illustrating an artificial intelligence feedback method according to one embodiment. 2 to 7 are diagrams illustrating an artificial intelligence feedback method according to an embodiment.

[0029] First, referring to FIG. 1 and FIG. 2, input 1 and input 2 are provided to a script extractor and executor (S100). Specifically, referring to (1) and (2) in Figure 2, input 1 containing basic material for the artificial intelligence to learn and input 2 containing core content for learning are provided to the script extractor and executor.

[0030] As an example, input 1 may include multiple image files and input 2 may include, but is not limited to, edge intensities. As another example, input 1 may include a plurality of document files, and input 2 may include, but is not limited to, a set of words.

[0031] This is the stage where input 1 and input 2 are presented, where the material that cognitive intelligence needs to learn or train is prepared. It consists of the basic materials for artificial intelligence to learn, input 1 is document files (text information such as pdf, txt files) and image files (image information such as jpg, png files), and input 2 is the keywords or core content that the artificial intelligence must learn.

[0032] As an example, if input 1 is a document file, input 2 is the key subject words, and if input 1 is an image file, input 2 is the light and dark gradations or contours of the image. Next, the script extractor and executor provides the artificial intelligence with an initial condition for the popularity frequency value or rank value of information related to input 2 from input 1 (S200).

[0033] Specifically, referring to (3) in FIG. 2, the script extractor and executor provides the artificial intelligence with initial conditions for the popularity frequency value or ranking value of information related to input 2 from input 1. Script extractors and executors include programs that extract and execute scripts, such as Perl Script, java, or language script analyzers.

[0034] The script extractor and executor provides the recognition intelligence with initial conditions for the popularity frequency values ​​or ranking values ​​of information related to key subject terms and core content based on the content of input 2 in the initial learning material.

[0035] Next, the artificial intelligence generates a two-dimensional information table A in the form of a matrix with the keywords and core information based on the popularity frequency value or ranking value derived from the initial conditions as the two axes (S300).

[0036] Specifically, referring to (1) and (2) in Figure 2, the artificial intelligence generates a two-dimensional information table A in the form of a matrix with the axes representing keywords and core information based on popularity frequency values ​​or ranking values ​​derived from the initial conditions provided by the script extraction and execution device. The artificial intelligence prepares to create an initial two-dimensional information table A with information related to initial popularity frequency values ​​or ranking values, etc.

[0037] The artificial intelligence generates a two-dimensional information table A, which is a two-dimensional rectangular table coordinate system, with the key subject words and the core contents found from the document file or image file according to the initial popularity frequency value or ranking value. The two-dimensional information table A has a structure similar to that of a matrix, and may be a two-dimensional rectangular table coordinate system with the keywords and the core information on the x-axis and y-axis, respectively.

[0038] Next, through interaction between the artificial intelligence and the script extraction and execution unit, a popularity frequency value or a ranking value is added to the corresponding coordinates of the two-dimensional information table A, and the knowledge acquired by the artificial intelligence is matched to the corresponding coordinates (S400).

[0039] Specifically, referring to (3) in Figure 2, through interaction between the artificial intelligence and the script extraction and execution device, a popularity frequency value or ranking value is added to the corresponding coordinates of the two-dimensional information table A, and the knowledge acquired by the artificial intelligence is matched to the corresponding coordinates.

[0040] Through the interaction between the AI ​​and the script extraction and execution device, popularity frequency, weighting, or ranking value can be added to the corresponding coordinates of the two-dimensional information table, which is a two-dimensional rectangular table coordinate system, and the knowledge acquired by the AI ​​can be matched and managed at the corresponding coordinates.

[0041] Next, referring to Figures 2 and 3, the artificial intelligence digitizes the ranking values ​​of the two-dimensional information table A from the two-dimensional information table A, and each coordinate is randomly turned on or off to generate a two-dimensional pseudo-linear conversion table T having the same area as the two-dimensional information table A (S500).

[0042] Specifically, referring to (5) of Figure 2 and Figure 3, the artificial intelligence digitizes the ranking values ​​of the two-dimensional information table A from the two-dimensional information table A, and generates a two-dimensional pseudo-linear conversion table T having the same area as the two-dimensional information table A, in which each coordinate is randomly turned on or off.

[0043] An artificial intelligence system generates a two-dimensional pseudo-linear conversion table T having the same size as the two-dimensional information table A. The artificial intelligence projects the ranking values ​​for the information in the two-dimensional information table A directly onto the points corresponding to each coordinate in the two-dimensional pseudo-linear conversion table T, digitizing them, and generates a two-dimensional pseudo-linear conversion table T in which the digitized values ​​of each coordinate are maintained as they are, but each coordinate has a random on and off state that is unpredictable and has no special pattern. The two-dimensional pseudo-linear conversion table T generated by the artificial intelligence is a calculation table for performing pseudo-linear conversion.

[0044] Referring to FIG. 3, it shows how the artificial intelligence generates a 2D pseudo-linear conversion table for each step used in the calculation in the feedback process and the calculation-based derivation process, and how the shape is constructed.

[0045] The two-dimensional pseudo-linear conversion table for each stage is configured to maintain the ranking values ​​as they are, and must be in a random on and off state in the same coordinate system. This means that each coordinate is randomly arranged to be 1 or 0, and the two-dimensional pseudo-linear conversion table is created by randomly generating the values ​​in an unpredictable manner. As an example, the left and right tables of FIG. 3 can be multiplied for each same coordinate to generate random on and off states of a two-dimensional pseudo-linear transformation table.

[0046] Next, referring to Figures 4 and 5, the artificial intelligence performs pseudo-linear transformation multiple times on the two-dimensional information table A to calculate a two-dimensional pseudo-linear transformation table T, thereby forming a two-dimensional inherent characteristic table in which each coordinate has a characteristic vector, and derives information of coordinates among the coordinates in the two-dimensional inherent characteristic table whose characteristic vectors have not been changed as deep truth values ​​(S600).

[0047] Specifically, referring to (6), (7), and (8) of Figure 4 and Figure 5, the artificial intelligence performs pseudo-linear transformation multiple times to calculate a two-dimensional pseudo-linear transformation table T on the two-dimensional information table A, to form a two-dimensional inherent characteristic table in which each coordinate has a characteristic vector, and derives information of the coordinates whose characteristic vectors have not been changed among the coordinates in the two-dimensional inherent characteristic table as deep truth values.

[0048] When the two-dimensional pseudo-linear conversion table T is operated n times on the two-dimensional information table A, an effect similar to a matrix operation appears. In the process of computing the two-dimensional pseudo-linear conversion table T on the two-dimensional information table A multiple times, the two-dimensional information table A is converted into a two-dimensional inherent characteristic table having a characteristic vector, and at this time, the information that remains unchanged until the end is derived as a deep truth value that does not substantially change.

[0049] When the two-dimensional pseudo-linear transformation table T is operated n times on the two-dimensional information table A, n pseudo-linear transformations are performed. The two-dimensional pseudo-linear conversion table T generated here is generated by multiplying factors at the same position, unlike general matrix operations, and each coordinate is turned on and off randomly.

[0050] On the other hand, a two-dimensional pseudo-linear conversion table and a rectangular table of the same area that randomly provides on and off states can be calculated to allow arbitrary rotation of 90 degrees, 180 degrees, 270 degrees, or 360 degrees, and the truth value can be derived by calculation. Referring to (6) of FIG. 4, it can be seen that the pseudo-linear transformation is performed n times.

[0051] Referring to (7) of FIG. 4, the pseudo-linearly transformed two-dimensional characteristic table values ​​are composed of characteristic vectors and characteristic values ​​that do not fluctuate any more or are closest to the truth. This is done by finding only the necessary information coordinate system so that only those that can be said to be true survive in the initial two-dimensional information table coordinate values, and the information that matches the surviving coordinate values ​​at this time becomes a deep truth value.

[0052] Referring to (8) of Figure 4, the process of deriving deep truth values ​​is similar to finding eigenvalues ​​and eigenvectors in a matrix. However, unlike matrix calculation, pseudo-linear transformation is performed in the form of one-to-one multiplication of a two-dimensional pseudo-linear transformation table of the same area, in which each coordinate value is assigned an eigenrank value that is randomly on and off, by the same coordinate system of the two-dimensional information table for each stage. This makes it different from matrix calculation.

[0053] Referring to FIG. 5, the entire process shown in (6), (7), and (8) of FIG. 4 can be viewed as a continuous combination of T-table calculation processes for the same-dimensional pseudo-linear transformation of a two-dimensional coordinate table consisting of the coordinate system of the original two-dimensional information table and its corresponding coordinate values, and S-table calculations for characteristic table calculations to find characteristic values ​​or deep truth values.

[0054] This paper conceptually shows how a high proportion of the artificial intelligence knowledge and learning content that is matched with each coordinate value in such a two-dimensional coordinate table survives in the coordinate system and can be derived as deep truth values. The n pseudo-linear transformations result in a unique rank value being assigned to each coordinate, or not, depending on the random on / off state of that coordinate.

[0055] To explain this in detail again, each coordinate switches each time and the coordinate value is either on or off. When this happens, only the coordinate values ​​for the on state are maintained as they are, and the coordinate values ​​for the off state are all processed to 0. The coordinate values ​​for the off state no longer have any effect on the corresponding coordinates in the 2D information table to which the knowledge was matched in the relevant calculation, and the knowledge of that coordinate is recognized as unimportant, and only the coordinates of the on state are recognized as important information in proportion to the weight of their ranking value.

[0056] In this way, the on / off state changes randomly each time n pseudo-linear transformations are performed, acting like an unpredictable popularity vote. If the target survives to the end and has a relatively high weight and ranking value, the unique characteristic table and unique characteristic value are selected by calculation according to the unique characteristic table, reflecting this state.

[0057] This works as if determining the eigenvalues ​​and eigenvectors of a matrix to determine the order, but the calculation itself goes through a different process from the calculation of a matrix, since the conversion table and the knowledge information table used in the calculation are tables generated by performing calculations at the same position. That is, the process of finding the unique property table is unique.

[0058] According to the above-described AI feedback method of the embodiment, the AI ​​learns the ultimate deep truth value from input 1 and input 2. In addition, the two-dimensional inherent characteristic table converted from the two-dimensional information table by the artificial intelligence feedback method according to an embodiment generates a first generation information table layer.

[0059] Through the artificial intelligence feedback method and through the process of the artificial intelligence learning from other information and finding deep truth values, the two-dimensional information table layers of the second generation, third generation, fourth generation, and nth generation are generated respectively. The artificial intelligence can convert the (n×n) type two-dimensional coordinate system into a linear one having (n×n) channels for easier comparison and analysis using the generated n-generation two-dimensional information table layer, and can perform comparison and analysis, and using this, the two-dimensional coordinate system can be mapped linearly.

[0060] Next, referring to Figures 6 and 7, the artificial intelligence converts the two-dimensional inherent characteristic table into multiple generations of one-dimensional linear channels (1st generation knowledge layer to nth generation knowledge layer) to which deep truth values ​​are mapped and stores them in a repository, and organizes the knowledge context formed by connecting the deep truth values ​​of each of the multiple generations of one-dimensional linear channels (1st generation knowledge layer to nth generation knowledge layer) into categories (knowledge category 1 to knowledge category n) and stores them in the repository (S700).

[0061] Specifically, referring to FIG. 6, the artificial intelligence uses a repository to convert all two-dimensional coordinates of the two-dimensional unique characteristic table into multiple generations of one-dimensional linear channels (first generation knowledge layer to nth generation knowledge layer) having linear channel values, maps and fills deep truth values ​​into the corresponding channels, and connects the deep truth values ​​of each of the multiple generations of one-dimensional linear channels to form a context of information and knowledge.

[0062] As an example, if the two-dimensional inherent characteristic table composed of deep truth values ​​acquired by the artificial intelligence from each generation is a (n×n) type two-dimensional coordinate system, each coordinate can be mapped to linear (n×n) channels, and the corresponding channel represented by the red bar with the deep truth value for each generation in Figure 6 is connected to the value of the knowledge layer with the deep truth value related to the next generation. By linking multiple knowledge layers in this way, knowledge threads can be inferred and formed.

[0063] Referring to Figure 7, the artificial intelligence constructs systematized knowledge by organizing and storing the main information that constitutes the context of information and knowledge in the repository by category (knowledge category 1 to knowledge category n). For example, artificial intelligence can use a repository to organize knowledge built on deep truth values ​​or deep knowledge contexts linked over several generations into categories of knowledge.

[0064] In one embodiment, the artificial intelligence feedback method constructs a coordinate system based on initial input data and input 1 and input 2 based on core keywords, and each coordinate value is assigned a relative ranking value based on frequency, priority, etc., and knowledge found by the artificial intelligence is matched and accumulated with the coordinate system and coordinate values ​​constructed in a two-dimensional coordinate table, thereby efficiently accumulating knowledge from the beginning.

[0065] Therefore, the artificial intelligence feedback method according to one embodiment is a more efficient method of deriving deep truth values, which is the ultimate goal, and reaching a final conclusion than conventional methods, and by conceptually realizing this and processing and analyzing information, it can help the artificial intelligence to reach a final conclusion more efficiently, as well as being used as a core technology based on knowledge construction of artificial intelligence, thereby increasing its utility.

[0066] That is, an artificial intelligence feedback method is provided that efficiently provides intuitive feedback for learning. Hereinafter, an artificial intelligence feedback system according to another embodiment will be described with reference to FIGS. The artificial intelligence feedback system according to another embodiment may perform the artificial intelligence feedback method according to the above-mentioned embodiment, but is not limited thereto.

[0067] FIG. 8 is a diagram illustrating an artificial intelligence feedback system according to another embodiment. Referring to FIG. 8, an artificial intelligence feedback system according to another embodiment includes a script extractor and executor 100, an artificial intelligence 200, an output unit 300, and a repository 400. The script extractor and executor 100 receives an input 1 including basic material for learning and an input 2 including core content for learning, and generates initial conditions for popularity frequency values ​​or ranking values ​​of information related to the input 1 to the input 2.

[0068] The artificial intelligence 200 is connected to the script extractor and executor 100, and generates a two-dimensional information table in the form of a matrix with keywords and core information based on popularity frequency values ​​or ranking values ​​derived from initial conditions as the two axes, adds popularity frequency values ​​or ranking values ​​to corresponding coordinates of the two-dimensional information table, and matches knowledge acquired by the artificial intelligence to the corresponding coordinates. The artificial intelligence 200 generates a two-dimensional pseudo-linear conversion table having the same area as the two-dimensional information table, in which the ranking values ​​of the two-dimensional information table are digitized from the two-dimensional information table and each coordinate is randomly turned on or off, performs pseudo-linear conversion on the two-dimensional information table multiple times to calculate the two-dimensional pseudo-linear conversion table, forms a two-dimensional inherent characteristic table in which each coordinate has a characteristic vector, derives information of coordinates whose characteristic vectors have not been changed among the coordinates of the two-dimensional inherent characteristic table as deep truth values, converts the two-dimensional inherent characteristic table into multiple generations of one-dimensional linear channels to which deep truth values ​​are mapped and stores them in the repository 400, and organizes the context of knowledge formed by connecting the deep truth values ​​of each of the multiple generations of one-dimensional linear channels by category and stores them in the repository 400.

[0069] The output unit 300 is connected to the artificial intelligence 200 and outputs the deep truth value that the artificial intelligence 200 has derived from the two-dimensional information table. The repository 400 is connected to the artificial intelligence 200, and stores the context of knowledge formed by connecting deep truth values ​​derived by the artificial intelligence 200 by category.

[0070] 9 and 10 are diagrams for explaining an artificial intelligence feedback system according to another embodiment. Referring to Figures 8 and 9, an artificial intelligence recognition program is shown with input 1, which is information relating to a document or image corresponding to the initial input from which to learn, and input 2, which corresponds to priority keywords that should be treated preferentially. The recognition program, which represents an artificial intelligence, interacts with a script extractor and executor that provides a means to interact or examine input data. The script extractor and executor is a powerful system language such as a Perl script that searches for information in documents or input files, extracts data required by the recognizer, and utilizes the recognizer in calculating frequency values. The Perl script can perform functions such as word extraction, adjacent word extraction, basic searching and statistics on sentence similarity.

[0071] The artificial intelligence collects and feeds back analytical information through a special window of such built-in script grammar, and stores it in a two-dimensional information table that converts specific information into knowledge, or stores the learned results by matching them with the coordinates of a two-dimensional coordinate table. The two-dimensional coordinate table can be based on coordinates that serve as information coordinate axes represented by input 1 and input 2. At this time, the information coordinate axis can be converted into a vertical line shape according to the convenience of calculation. The stop command sets the initial condition for how many times the artificial intelligence should perform learning. In particular, when no initial conditions are given, the artificial intelligence (AI) that is the recognition program learns through the process of n pseudo-linear transformations until the values ​​in the characteristic vector table or characteristic table hardly change, as if it were searching for the eigenvalues ​​and eigenvectors of the matrix and finding a deep truth value that does not require any further feedback or calculations.

[0072] If there is no special input for the stop command, it operates in this manner, and when a stop command is given, the learning and calculation of the artificial intelligence ends with the stop command, and the deep truth value, which is the learning result, is output to the output section. At this time, the AI ​​can organize the knowledge context formed by connecting deep truth values ​​into categories and store them in a repository, thereby building systematized knowledge.

[0073] Referring to FIG. 10, there is shown a conceptual diagram of how the extraction of relevant scripts, the execution of the scripts, and the artificial intelligence recognition program interact. The recognition program, which is an artificial intelligence, can ensure that it has a structure that searches for and analyzes the required information and words by manipulating the scripting language.

[0074] In this way, the recognition program uses a script extractor and executor to continuously perform feedback on itself using input 1 and input 2, creates a two-dimensional information table and a two-dimensional table coordinate system, creates a coordinate distribution in the form of popularity frequency value or ranking value as the weight for knowledge, and stores specific sentence information for those coordinates separately in a coordinate system of the same size, and information matching those coordinates is continuously accumulated and stored. All the above information can be stored in a repository by artificial intelligence.

[0075] Hereinafter, with reference to FIG. 11, image-based training will be described as an example of learning using the AI ​​feedback method and AI system. FIG. 11 is an image showing training by images using an artificial intelligence feedback method and an artificial intelligence system.

[0076] Referring to FIG. 11, image training can be a process of finding the features of jpg files, deriving deep truths, finding comprehensive commonalities, and generating diagrams. Specifically, jpg files having pixels of a certain size are trained. For the characteristics of image files, the popular frequency value (weighted value of variables and similar concepts) is given large around the contour line, and the assumption is made that the rest are not distinctive. Since an image file can also be viewed as a kind of document, a recognition program recognizes the value of each pixel in the image file, calculates the popularity frequency value by itself, and sets an accumulated value for the corresponding pixel. After a certain period of time has passed, a characteristic value is given according to the cumulative distribution value, and the rank values ​​are re-ranked according to the rank of this characteristic value. Here, the ranking value acts like a weighting value.

[0077] Each pixel can be matched to each coordinate of a table on a certain plane, i.e., a matrix-like table, and the matrix-shaped table has a distribution of information values ​​related to the image. From these values, the ranking value of each pixel for the information value of the matrix-shaped table created by the recognition program is maintained as it is, but a matrix-shaped table of the same dimension is created to create a random table that shows or hides the ranking value for each coordinate of each table, and this table is operated like a linear transformation of the matrix to find the deep truth value.

[0078] When the script executes a pseudo-linear transformation of each random table, each transformed table produces a deep truth property vector that behaves in the same way as the eigenvalues ​​and eigenvectors of a matrix. Once the feature vector values ​​are generated, they can be ranked according to the magnitude of the feature values, which serves to highlight pixels in the image that exhibit the image's features, completing a kind of deep truth finding process. The final result is a standard image that is completed by finding a kind of commonality between all the images it has learned.

[0079] Hereinafter, with reference to FIG. 12, document-based training will be described as another example of learning using an artificial intelligence feedback method and an artificial intelligence system. FIG. 12 is an image showing the artificial intelligence feedback method and written training using an artificial intelligence system.

[0080] Referring to FIG. 12, document training can be the process of finding relevance and creating deeper truth from txt documents. Specifically, a document having a certain size, in this case a txt file, is selected and trained. The learning process involves inputting a txt file into input 1, and inputting keywords or topics of interest into input 2. With the input 2 suggested words or words of interest, the popularity frequency distribution of the keywords and sentences containing the keywords are found and characteristic values ​​are given based on the frequency.

[0081] Keywords are matched to a matrix-shaped table to create a layer of knowledge to be learned, and their frequency distribution is matched to coordinate values ​​to determine which coordinates they correspond to. In the embodiment, approximately seven keywords are listed, and a table similar to a 7x7 matrix taking into account combinations of three words is constructed, and coordinate values ​​are generated with related words as axes (represented as numbers in the coordinate system). Each coordinate is assigned a ranking value based on its popularity frequency, and after the ranking value is assigned to each coordinate, the value of each coordinate is recognized by a recognition program, which calculates the popularity frequency value itself and sets an accumulated value for the corresponding coordinate.

[0082] After a certain period of time has passed, a characteristic value is given according to the cumulative distribution value, and the rank values ​​are re-ranked according to the rank of this characteristic value. Here, the ranking values ​​act similarly to the weighting values. Each coordinate can be matched with each coordinate of a table on a certain plane, i.e., a matrix-like table, and the matrix-shaped table has a distribution of information values ​​of each coordinate. From these values, the ranking value of each coordinate for the information value of the matrix-shaped table created by the recognition program is maintained as it is, but a matrix-shaped table of the same dimension is created to create a random table that displays or hides the ranking value for each coordinate of each table, and this table is operated like a linear transformation of the matrix to find the deep truth value.

[0083] After a certain period of time has passed, the ranking values ​​of each coordinate in the table are maintained as they are, but a characteristic vector representing the values ​​is randomly generated, which behaves in the same way as the eigenvalues ​​and eigenvectors of a matrix. Once the feature vector values ​​are generated, they can be ranked according to the magnitude of the feature values ​​of the feature vectors, which highlights meaningful sentences containing keywords in the txt or document, completing the process of finding a kind of deep truth value through combination or fusion. The final result will have a sort of overall conclusion drawn from the studied documents.

[0084] FIG. 13 is a diagram for explaining an artificial intelligence feedback system according to another embodiment based on FIG. Based on the above and referring to Figure 13, a method for implementing the artificial intelligence feedback system includes a method of having a separate repository and sharing accumulated knowledge through successive generations by utilizing the repository, and linking knowledge that has a large influence between generations to create a knowledge tradition and context.

[0085] The present invention maintains the functionality of the artificial intelligence feedback system while adding another process to obtain more enhanced feedback and deeper truth values. Thus, the method pursued in this invention requires 10 steps. The added process is to make the table calculation method work in a diverse manner. In order to maintain the truth value not being 0, the additional process of the present invention is to create an operation table with a non-zero but relatively small value as a substitute value for 0 for the continuity of table operation. In the present invention, a method of replacing a part that becomes 0 with 0.1 is also added as an example.

[0086] In this way, we create a structure that responds to any operational transformation and obtains deeper truth values ​​as the number of operations increases, regardless of whether we rotate the orientation of the table acting as an operator, flip it 180 degrees, use symmetry about a central point, or change the front and back sides.

[0087] The first stage is when a recognition program (artificial intelligence) is presented with input 1 and input 2, which provide the material to learn or train. It consists of basic materials for artificial intelligence to learn, input 1 may be mainly document files (such as text information like pdf, txt files) and image files (such as image-like information like jpg, png files), and input 2 may be mainly keywords or core content that artificial intelligence should learn. For example, in the case of a document file it is mainly key subject words, in an image file it is the shades of light or the contours of a picture.

[0088] The second stage is the script extraction and execution stage, in which initial conditions such as popularity frequency values ​​or ranking values ​​of information related to key subject terms and core content based on the contents of input 2 in the initial learning material are given to the recognition program (artificial intelligence). The third step is a step in which the recognition program operates, in which the recognition program prepares to create an initial two-dimensional information table with information related to initial popularity frequency values ​​or ranking values.

[0089] The fourth step is when the recognition program (artificial intelligence) creates a two-dimensional information table. In this stage, a two-dimensional rectangular table coordinate system is created with the key subject terms and the core contents found from the document / image files according to the initial popularity frequency values ​​or ranking values. In other words, this is called a two-dimensional information table. The shape has a structure similar to what we have come to know as a matrix, but it is a two-dimensional table coordinate system with keywords and core information as axes.

[0090] The fifth step is to assign the popularity frequency, weight, or ranking value of the corresponding coordinate value to the 2D table coordinate system through the interaction between the artificial intelligence and the script extractor / executor, and to match and manage the knowledge acquired by the artificial intelligence to the corresponding coordinate.

[0091] The sixth step is to create a two-dimensional table of the same size as the two-dimensional information table, project the ranking values ​​for the information in the two-dimensional information table onto the points corresponding to each coordinate, and digitize them. The digitized values ​​of each coordinate are maintained as they are, but each coordinate has no special pattern and is in a random on / off state that is unpredictable, creating a two-dimensional table. The two-dimensional table generated at this stage is a calculation table for performing pseudo-linear transformation.

[0092] The seventh step is to derive the results obtained by the pseudo-linear transformation. In this step, when multiple calculations are performed using a two-dimensional table for pseudo-linear conversion, this has an effect similar to a matrix calculation, but in this calculation process, the two-dimensional information table is converted into a characteristic information table having a characteristic vector, and at this time, the information that remains unchanged until the end is derived as a deep truth value that hardly changes.

[0093] Through this process, the artificial intelligence finds the ultimate deep truth value based on input 1 and input 2, and the two-dimensional information table thus formed constitutes the first generation information table layer. Similarly, the artificial intelligence learns from other information and finds deeper truth values, generating a second-generation information table layer, and then using the same method to generate third-generation, fourth-generation, and n-th-generation two-dimensional information table layers, respectively.

[0094] Eighth, in the multiple calculation steps using the two-dimensional table for pseudo-linear conversion, various conversions can be performed between steps to further strengthen the previously performed calculation steps. In this transformation process, as shown in FIG. 1, transformation operations are performed in any form, such as rotation, symmetry, point symmetry, axis symmetry, and inversion.

[0095] Here, the conversion operation replaces the existing table values ​​that were processed to 0 with very small values ​​of around 0.1. This serves to alleviate the phenomenon in which the surviving truth values ​​are relatively reduced when performing various conversion operations, when applying the accumulated value. By adding and inserting such various conversion operations, it is possible to derive stronger cumulative truth values ​​when generating the second generation information table, third generation, fourth generation, and nth generation two-dimensional information table layers, respectively.

[0096] The ninth step is the process in which the recognition program uses the repository to convert all the 2D coordinates of the 2D information table layer into linear channel values, maps and fills the deep truth values ​​into the corresponding linear channels, and connects the channels of different generations to create an information / knowledge context. The tenth step is the process of constructing systematized knowledge by dividing important information that constitutes the context of information / knowledge into knowledge categories by category in the repository.

[0097] 14 to 18 are diagrams illustrating an artificial intelligence feedback system according to still another embodiment based on FIG. Based on the above, reference is made to FIGS. 14 to 18. FIG. 14 shows the basic conceptual structure of a system for implementing the above-mentioned invention. FIG. 15 shows how the script extractor / executor and the recognizer interact with each other.

[0098] 16 and 17 show the working concept and detailed configuration of the artificial intelligence feedback system. FIG. 18 shows how deep truth values ​​are reached by using a two-dimensional coordinate table A that shows the distribution of weights or popularity frequency values ​​obtained by learning information based on popularity frequency values ​​or importance of artificial intelligence according to the coordinate axes of input 1 and input 2.

[0099] FIG. 19 shows the structure of a conversion table having unique rank values ​​acting as operators when performing pseudo-linear conversion for each time or each step, and shows the on / off state for each coordinate.

[0100] (1) and (2) in Figure 14 indicate "input 1," which is information relating to a document or image that corresponds to the initial input from which the recognition program (artificial intelligence) must learn, and "input 2," which corresponds to a priority keyword that should be treated with priority. (3) and (4) in FIG. 14 may be considered to be script extractors and / or script executors that provide a means for a recognition program (5), which corresponds to artificial intelligence, to interact with or examine input material. (3) and (4) are implemented using the same system language, Perl, which is a powerful language for searching for information in documents or input files, extracting data required by the recognition program, and calculating frequency values.

[0101] Perl scripts can perform functions such as word extraction, adjacent word extraction, and basic searching / statistics for sentence similarity. The artificial intelligence uses a special window of such built-in script grammar to collect and feed back analytical information, and stores it in a two-dimensional information table that converts specific information into knowledge, or in a two-dimensional coordinate table by matching the learned results with the coordinates of the two-dimensional coordinate table.

[0102] The two-dimensional coordinate table is based on coordinates that serve as the information coordinate axes represented by "Input 1" and "Input 2."

[0103] At this time, the information coordinate axis can be converted into a vertical line shape according to the convenience of calculation. (5) serves the function of providing initial conditions for how many times the artificial intelligence should learn. If no special initial conditions are given, the recognition program (5) goes through n pseudo-linear transformation processes, as if it were finding the eigenvalues ​​and eigenvectors of the matrix until the value of the characteristic vector table or characteristic table hardly changes, and learns until it finds a deep truth value that does not require any further feedback or calculations. It operates in this manner unless there is any special input for the stop command (6), and if a stop command is given, the artificial intelligence's learning and calculations end upon receipt of the stop command, and it outputs the learning result (7).

[0104] Figure 15 conceptually shows how the extraction of scripts corresponding to (3), (4), and (5), the execution of the scripts, and the recognition program (artificial intelligence) interact with each other in the basic conceptual diagram of Figure 14. As shown in Figure 15, the recognition program (artificial intelligence) is structured to search for and analyze the necessary information and words by manipulating the scripting language. In this way, the recognition program continuously feeds back to itself using the script extractor / executor with input 1 and input 2, generates a two-dimensional information table and a two-dimensional table coordinate system, and generates a coordinate distribution in the form of popularity frequency value or ranking value as the weight for knowledge. Specific sentence information related to the coordinates is stored separately in a coordinate system of the same size, and information matching the coordinates is continuously accumulated and stored.

[0105] 16 and 17 show in detail the implementation and working concept of the artificial intelligence feedback system. Figure 16 shows the process in which a script extractor / executor and recognition program (artificial intelligence) having input 1 and input 2 generates an initial two-dimensional information table through its own feedback and training, and manages the two-dimensional information table coordinate system by matching knowledge corresponding to each coordinate of the two-dimensional information table. The process of generating and managing a coordinate system based on input corresponds to (1), (2), (3), and (4) in FIG. 16. Through this process, a pseudo-linear transformation is performed for each coordinate system to find the true truth value, and in order to perform the pseudo-linear transformation, a linear transformation table of the same dimension is generated. The process of generating a conversion table of the same dimension for the pseudo-linear conversion corresponds to (5) in FIG.

[0106] FIG. 17 shows how the initial two-dimensional information / knowledge table is transformed with the two-dimensional pseudo-linear transformation table generated by the process of FIG. 16, and how the deeper truth values ​​are derived. It can be seen from (7) of Figure 17 that the pseudo-linear transformation is performed n times, and (8) of Figure 17 creates a two-dimensional table composed of characteristic vectors and characteristic values ​​so that the values ​​of the pseudo-linearly transformed two-dimensional table do not fluctuate any more or are closest to the truth. This involves finding only the necessary information coordinate system so that only those coordinate values ​​in the initial knowledge / information two-dimensional table that can be said to be true truths survive, and the information that matches the surviving coordinate values ​​at this time becomes deep truth values. The process of deriving deep truth values ​​is similar to finding eigenvalues ​​and eigenvectors using a matrix, but in the present invention, unlike matrix operations, each coordinate value of a two-dimensional pseudo-linear conversion table of the same size with eigenrank values ​​assigned is converted in a one-to-one manner by multiplying each coordinate value of a two-dimensional pseudo-linear conversion table for each stage, which is generated by creating an unpredictable random on / off state for each stage, with the same coordinate system, which is different from matrix operations. In the present invention, such pseudo-linear transformation is very distinctive.

[0107] FIG. 18 shows in detail what the pseudo-linear transformation means, based on the concepts of FIG. 16 and FIG. As shown in FIG. 18, the entire process of FIG. 17 can be seen as a continuous combination of the S table operation for characteristic table operation to find characteristic values ​​or deep truth values, in which the two-dimensional coordinate table consisting of the coordinate system of the original two-dimensional knowledge information table and its corresponding coordinate values ​​is organized by a continuous operation process of the T table for the same-dimensional pseudo-linear transformation. Figure 18 conceptually shows how a high percentage of the artificial intelligence knowledge and learning content that matches each coordinate value in such a two-dimensional coordinate table survives in the coordinate system and is derived as a deep truth value. The n pseudo-linear transformations result in a unique ranking value being assigned to each coordinate, or not, depending on the random on / off state of each coordinate. To explain this in detail again, each coordinate switches each time and the coordinate value is either on or off. When this happens, only the coordinate values ​​for the on state remain unchanged for each coordinate value, and the coordinate values ​​for the off state are all processed to 0. The coordinate values ​​for the off state no longer have any effect on the corresponding coordinate in the knowledge / information table in that calculation, and the knowledge of that coordinate is recognized as unimportant, and only for the coordinates in the on state, it is recognized as important information in proportion to the weight of its ranking value. In this way, the on / off state changes randomly each time n pseudo-linear transformations are performed, acting like an unpredictable popularity vote.

[0108] In this way, if a surviving criterion has a relatively high weight and ranking value, a unique characteristic table and unique characteristic value are selected by calculation according to the unique characteristic table, reflecting this state. This works as if the eigenvalues ​​and eigenvectors of a matrix are found and the order is determined, but the calculation itself goes through a different process from the calculation of a matrix because the conversion table and the knowledge information table used in the calculation are tables generated by performing calculations in the same position. Therefore, in the present invention, the process of finding the unique characteristic table is unique.

[0109] FIG. 19 shows how the recognition program of FIG. 18 generates the pseudo-linear conversion table for each step used in the feedback process and the derivation process by calculation, and how the shape is constructed. In FIG. 3, the pseudo-linear conversion table for each stage is configured to maintain the ranking value as it is. And it should be in a random on / off state in the same coordinate system, which is arranged so that each coordinate is randomly set to 1 or 0, but it is generated randomly so that it is unpredictable, creating a two-dimensional table. Then, the left table of FIG. 3 and the right table of FIG. 3 are multiplied for each same coordinate to generate a random on / off state of a pseudo-linear transformation two-dimensional table. In this way, if you proceed n times, you will perform n pseudo-linear transformations.

[0110] The pseudo-linear conversion table here is also generated by multiplying factors in the same position, which is different from general matrix operations. Using this method, the present invention can create a pseudo-linear conversion table that is efficiently switched on / off randomly. Additionally, to make the pattern and matching of the information table clearer, as can be seen from Figure 3, the on / off state of the two-dimensional table on the right side of Figure 3 can be changed to 1 for the on state and 0.1 for the off state as shown in Figure 19, and deep truth values ​​can be derived using all pseudo-linear transformations such as rotation, inversion, axial symmetry, point symmetry, and changing front and back.

[0111] As mentioned above, Figure 6 shows the process of mapping each coordinate value of the coordinate system of the two-dimensional knowledge / information table from which deep truth values ​​are derived to a linear channel, mapping the two-dimensional information / knowledge table layer for each generation or the knowledge layer for each generation up to n generations. To explain in more detail, the two-dimensional information / knowledge table composed of the deep truth values ​​acquired in each generation is identical to a two-dimensional lattice coordinate system. Therefore, if the two-dimensional coordinate system is, for example, an (n×n) type two-dimensional coordinate system, each coordinate can be mapped to a linear (n×n) number of channels, and the corresponding channel indicated by the red bar in Figure 7, which is assigned a deep truth value by generation, can be linked to the value of the knowledge layer of the next generation, which is assigned the associated deep truth value.

[0112] Therefore, by linking several layers of knowledge in this way, it is possible to infer the context of knowledge. Figure 7 explains how knowledge built up from deep truth values ​​or deep knowledge contexts linked over several generations is organized by categorizing the knowledge using a repository.

[0113] FIG. 20 shows the results of an artificial intelligence training process that at least in part reflects the processes of additional FIGS. As shown, by accumulating the conversion in various ways such as inversion, axial symmetry, point symmetry, and rotational symmetry, the conversion table can be converted in a variety of ways, resulting in clearer results than when the on / off values ​​are set to 1.0 and 0.1, respectively. In order to visualize the actual results, the process involves selecting only the values ​​that reach the maximum cumulative value under the condition that the intensity of the image contour is the popularity frequency value, and deriving them as the optimal training value. Additional training was performed with images to make the visualization relatively easy to follow.

[0114] Although an embodiment of the present invention has been described above with reference to the accompanying drawings, a person having ordinary skill in the art to which the present invention pertains can understand that the present invention can be embodied in other specific forms without changing its technical concept or essential features. Therefore, it should be understood that the above-described embodiment is illustrative in all respects and is not limiting.

Claims

1. An artificial intelligence feedback system including a script extractor and executor, an artificial intelligence coupled to the script extractor and executor for acquiring knowledge, an output unit coupled to the artificial intelligence, and a repository coupled to the artificial intelligence, The script extractor and executor receives an input 1 including basic material for the artificial intelligence to learn and an input 2 including core content for the artificial intelligence to learn, and generates an initial condition for a popularity frequency value that is a cumulative value of characteristic information related to the input 2, or a ranking value that is ranking information based on a distribution characteristic of the cumulative value, from the input 1; The artificial intelligence generates a two-dimensional information table in a matrix form with a keyword and a core information as axes, respectively, using the popularity frequency value or the rank value derived from the initial condition, adds the popularity frequency value or the rank value to a corresponding coordinate of the two-dimensional information table, and matches the knowledge to the corresponding coordinate; The output unit outputs a deep truth value derived by the artificial intelligence from the two-dimensional information table, The repository organizes and stores the context of knowledge formed by connecting the deep truth values ​​derived by the artificial intelligence by category, The artificial intelligence further comprises: The ranking values ​​of the two-dimensional information table are digitized from the two-dimensional information table, and each coordinate is randomly turned on or off to generate a two-dimensional pseudo-linear conversion table having the same area as the two-dimensional information table; A two-dimensional information table is subjected to a pseudo-linear transformation for calculating the two-dimensional pseudo-linear transformation table a plurality of times to form a two-dimensional inherent characteristic table in which each coordinate has a characteristic vector, and information of coordinates whose characteristic vectors are not changed among the coordinates of the two-dimensional inherent characteristic table is derived as the deep truth value; The two-dimensional characteristic table is converted into a plurality of generations of one-dimensional linear channels to which the deep truth values ​​are mapped, and the converted one-dimensional linear channels are stored in the repository. The deep truth values ​​of the plurality of generations of one-dimensional linear channels are linked to form a knowledge context, which is organized by category and stored in the repository. An artificial intelligence feedback system, wherein when the pseudo-linear transformation for calculating the two-dimensional pseudo-linear transformation table on the two-dimensional information table is performed a plurality of times, each time performing one of the following operational transformations: a rotation operation of the direction of the two-dimensional pseudo-linear transformation table, a 180-degree inversion operation, a symmetric operation about a center point, and a left-right symmetric operation.

2. 2. The artificial intelligence feedback system of claim 1, wherein when performing the pseudo-linear transformation to calculate the two-dimensional pseudo-linear transformation table on the two-dimensional information table multiple times, if the coordinate value of each coordinate is on or off, the on state is changed to 1 and the off state is changed to 0.1, and then one of the operational transformations is performed on the coordinates of the two-dimensional pseudo-linear transformation table.

3. said input 1 includes a plurality of image files; The artificial intelligence feedback system of claim 2 , wherein said input 2 comprises contour strength.

4. The input 1 includes a plurality of document files; The artificial intelligence feedback system of claim 2 , wherein the input 2 includes a set of words.

5. Providing an input 1 containing basic material for a knowledge-acquiring artificial intelligence to learn from, and an input 2 containing core content for the artificial intelligence to learn from, to a script extractor and executor; The script extractor and executor provides an initial condition for a popularity frequency value, which is a cumulative value of characteristic information related to the input 2, or a ranking value, which is ranking information based on a distribution characteristic of the cumulative value, to an artificial intelligence from the input 1; generating a two-dimensional information table in a matrix form with keywords and core information as axes on both sides by using the popularity frequency value or the rank value derived from the initial condition by the artificial intelligence; adding the popularity frequency value or the ranking value to the corresponding coordinates of the two-dimensional information table through interaction between the artificial intelligence and the script extractor and executor, and matching the knowledge to the corresponding coordinates; The artificial intelligence generates a two-dimensional pseudo-linear conversion table from the two-dimensional information table, in which the ranking values ​​of the two-dimensional information table are digitized and each coordinate is randomly turned on or off, and the two-dimensional pseudo-linear conversion table has the same area as the two-dimensional information table; The artificial intelligence performs a pseudo-linear transformation to operate the two-dimensional pseudo-linear transformation table on the two-dimensional information table a plurality of times to form a two-dimensional unique characteristic table in which each coordinate has a characteristic vector, and derives information of coordinates whose characteristic vectors are not changed among the coordinates of the two-dimensional unique characteristic table as deep truth values; The artificial intelligence converts the two-dimensional characteristic table into a plurality of generations of one-dimensional linear channels to which the deep truth values ​​are mapped, and stores the converted one-dimensional linear channels in a repository; and organizes a knowledge context formed by connecting the deep truth values ​​of each of the plurality of generations of one-dimensional linear channels into categories and stores the categories in the repository; An artificial intelligence feedback method, in which when the pseudo-linear transformation for calculating the two-dimensional pseudo-linear transformation table on the two-dimensional information table is performed multiple times, one of a rotation operation of the direction of the two-dimensional pseudo-linear transformation table, a 180-degree inversion operation, a symmetric operation about a center point, and a left-right symmetric operation is performed each time.

6. 6. The artificial intelligence feedback method according to claim 5, wherein when performing the pseudo-linear transformation for calculating the two-dimensional pseudo-linear transformation table on the two-dimensional information table multiple times, if the coordinate value of each coordinate is in an on or off state, the on state is changed to 1, and the off state is changed to 0.1, and then the coordinate of the two-dimensional pseudo-linear transformation table is subjected to one of the calculation transformations.

Images