CN111868683A - Operation implementation method and device in artificial intelligence application building and machine equipment - Google Patents
Operation implementation method and device in artificial intelligence application building and machine equipment Download PDFInfo
- Publication number
- CN111868683A CN111868683A CN201880002680.7A CN201880002680A CN111868683A CN 111868683 A CN111868683 A CN 111868683A CN 201880002680 A CN201880002680 A CN 201880002680A CN 111868683 A CN111868683 A CN 111868683A
- Authority
- CN
- China
- Prior art keywords
- artificial intelligence
- mathematical
- user
- intelligence application
- dictionary
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013473 artificial intelligence Methods 0.000 title claims abstract description 515
- 238000000034 method Methods 0.000 title claims abstract description 80
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 174
- 238000010276 construction Methods 0.000 claims description 50
- 238000012549 training Methods 0.000 claims description 28
- 238000012545 processing Methods 0.000 claims description 15
- 238000013519 translation Methods 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 35
- 238000010586 diagram Methods 0.000 description 22
- 230000009471 action Effects 0.000 description 14
- 238000005516 engineering process Methods 0.000 description 12
- 238000006243 chemical reaction Methods 0.000 description 10
- 238000004891 communication Methods 0.000 description 10
- 238000013527 convolutional neural network Methods 0.000 description 9
- 238000011161 development Methods 0.000 description 9
- 230000018109 developmental process Effects 0.000 description 8
- 230000001960 triggered effect Effects 0.000 description 8
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000004806 packaging method and process Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 230000003993 interaction Effects 0.000 description 5
- 230000014509 gene expression Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 230000005236 sound signal Effects 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000005538 encapsulation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000008140 language development Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Stored Programmes (AREA)
Abstract
An operation implementation method, device and machine equipment in artificial intelligence application building, wherein the method comprises the following steps: receiving an artificial intelligence algorithm configuration selection of the constructed artificial intelligence application by a user, wherein the artificial intelligence algorithm configuration is a network topology describing mathematical operations executed in the artificial intelligence application (310); selecting a dictionary describing the performed mathematical operation according to the artificial intelligence algorithm configuration (330); reconstructing mathematical operations on said dictionary by means of a mathematical language representation of graph theory, resulting in an executable text (350) performing said mathematical operations; and through the execution of the executable text, the user is enabled to run the built artificial intelligence application (370) according to the selected artificial intelligence algorithm configuration. The method, the device and the machine equipment can realize the translation of the algorithm graph obtained by the graphical programming into the executable artificial intelligence application, thereby achieving the purpose of running the artificial intelligence application.
Description
The invention relates to the technical field of Internet application, in particular to a method, a device and machine equipment for realizing operation in artificial intelligence application construction.
With the development of artificial intelligence technology, more and more artificial intelligence applications are making various decisions for obtained data based on artificial intelligence technology in many scenarios. Users have different artificial intelligence application requirements in different scenes, and therefore different artificial intelligence applications are often required to be searched for in different scenes.
However, in many cases, various artificial intelligence application requirements of users exist in a personalized manner, and many artificial intelligence applications released in the internet are not suitable.
At this time, with the rise of graphical programming, people gradually think that the graphical programming may be a path meeting the personalized requirements of the artificial intelligence application, and the graphical programming built by simple application is evolved into the implementation built by the artificial intelligence application.
However, the artificial intelligence application relates to a complex algorithm, and has very high programming difficulty, and it is practically difficult to obtain the executable artificial intelligence application through the algorithm graph built by the user on the graphical programming interface. In the process, how to translate the constructed algorithm graph to realize the operation of the corresponding artificial intelligence application is a dilemma for carrying out graphical programming by the artificial intelligence technology.
Disclosure of Invention
In order to solve the technical problem of how to translate an algorithm graph obtained by graphical programming into an executable artificial intelligence application and further realize operation in the related technology, the invention provides an operation realization method, an operation realization device and machine equipment in artificial intelligence application construction.
An operation implementation method in artificial intelligence application building, the method comprising:
receiving artificial intelligence algorithm configuration selection of a user on the constructed artificial intelligence application, wherein the artificial intelligence algorithm configuration is a network topology for describing mathematical operations executed in the artificial intelligence application;
selecting and obtaining a dictionary describing the executed mathematical operation according to the artificial intelligence algorithm configuration;
reconstructing mathematical operation on the dictionary through mathematical language representation of graph theory to obtain executable text for executing the mathematical operation;
and through the execution of the executable text, the user is enabled to run the built artificial intelligence application according to the selected artificial intelligence algorithm configuration.
An operation implementation device in artificial intelligence application building, the device comprising:
the receiving module is used for receiving artificial intelligence algorithm configuration selection of a user on the constructed artificial intelligence application, and the artificial intelligence algorithm configuration is a network topology for describing mathematical operations executed in the artificial intelligence application;
the dictionary acquisition module is used for selecting and acquiring a dictionary describing the executed mathematical operation according to the artificial intelligence algorithm configuration;
the text generation module is used for reconstructing mathematical operation on the dictionary through mathematical language representation of graph theory to obtain an executable text for executing the mathematical operation;
and the application running module is used for enabling the user to run the built artificial intelligence application according to the selected artificial intelligence algorithm configuration through the execution of the executable text.
A machine device, comprising:
a processor; and
a memory having computer readable instructions stored thereon which, when executed by the processor, implement a method according to the foregoing.
The technical scheme provided by the embodiment of the invention can have the following beneficial effects:
in the graphical programming for the artificial intelligence application, the artificial intelligence algorithm configuration selection for building the artificial intelligence application is obtained, the artificial intelligence algorithm configuration is a network topology for describing the mathematical operation executed in the artificial intelligence application, at the moment, a dictionary for describing the executed mathematical operation is obtained according to the artificial intelligence algorithm configuration selection, then the mathematical operation is rebuilt on the dictionary through the mathematical language representation of graph theory, an executable text for executing the mathematical operation is obtained, and finally, the operation of the built artificial intelligence application can be obtained by a user according to the selected artificial intelligence algorithm configuration through the execution of the executable text, so that the algorithm graph obtained by the graphical programming is translated into the executable artificial intelligence application, and the operation purpose is achieved.
According to the embodiment of the invention, the realization of translating the artificial intelligence application into the executable code is provided for the graphical programming of the artificial intelligence application, the possibility of realizing landing is further enhanced for graphically establishing the artificial intelligence application, the complexity of the artificial intelligence technology is not limited, and the artificial intelligence application can be established and operated in a personalized manner according to various artificial intelligence application requirements.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
FIG. 1 is a schematic illustration of an implementation environment in accordance with the present invention;
FIG. 2 is a block diagram illustrating an apparatus in accordance with an exemplary embodiment;
FIG. 3 is a flow diagram illustrating a method for implementing operations in the construction of an artificial intelligence application, in accordance with an illustrative embodiment;
FIG. 4 is a flowchart illustrating a description of step 350 according to a corresponding embodiment of FIG. 3;
FIG. 5 is a diagram illustrating the structure of a DAG corresponding to a simple mathematical primitive in accordance with an illustrative embodiment;
FIG. 6 is a flowchart illustrating a description of step 370, according to a corresponding embodiment of FIG. 3;
FIG. 7 is a flowchart illustrating a description of step 370 in another exemplary embodiment according to the corresponding embodiment of FIG. 3;
FIG. 8 is a flow diagram illustrating a method for implementing operations in the construction of an artificial intelligence application, in accordance with another illustrative embodiment;
FIG. 9 is a diagrammatic illustration of a graphical interface shown in accordance with an exemplary embodiment;
FIG. 10 is a diagram illustrating a block on an operator interface corresponding to a convolutional neural network operation, in accordance with an illustrative embodiment;
FIG. 11 is a schematic diagram of chunk distribution and linking of a three-layer convolutional neural network on an operator interface, according to the corresponding embodiment of FIG. 10;
FIG. 12 is a schematic diagram illustrating interactions between a front end and a back end in accordance with the present invention, according to an illustrative embodiment;
FIG. 13 illustrates a schematic diagram of an implementation of the overall scheme of the present invention in an exemplary embodiment;
FIG. 14 is a block diagram illustrating an apparatus for implementing operations in the construction of an artificial intelligence application, according to an example embodiment.
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the invention, as detailed in the appended claims.
The present invention relates to an implementation environment that, in one exemplary embodiment, includes at least a user terminal for providing a graphical programming interface. The user terminal may be various types of terminal devices, such as a smart phone, a tablet computer, a notebook computer, a desktop computer, and the like. Any user can realize the graphical programming related to the artificial intelligence application through the held user terminal, so that the artificial intelligence application required by the user can be obtained, and the artificial intelligence application requirements of the individual people can be met.
In the implementation environment, the user can freely build the needed artificial intelligence application, the operation of the built artificial intelligence application can be ensured, and the self-development of the artificial intelligence application is realized according to the self requirement.
It can be understood that the realization of the invention provides possibility for the construction of artificial intelligence application at any time by users, and no matter enterprise users or terminal users, the artificial intelligence application does not need to be autonomously developed with very large cost for self requirements.
FIG. 2 is a block diagram illustrating an apparatus according to an example embodiment. The apparatus 200 may be, for example, the user terminal 110 in the implementation environment shown in fig. 1.
Referring to fig. 2, the apparatus 200 includes at least the following components: a processing component 202, a memory 204, a power component 206, a multimedia component 208, an audio component 210, a sensor component 214, and a communication component 216.
The processing component 202 generally controls overall operation of the device 200, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations, among others. The processing components 202 include at least one or more processors 218 to execute instructions to perform all or a portion of the steps of the methods described below. Further, the processing component 202 includes at least one or more modules that facilitate interaction between the processing component 202 and other components. For example, the processing component 202 can include a multimedia module to facilitate interaction between the multimedia component 208 and the processing component 202.
The memory 204 is configured to store various types of data to support operations at the apparatus 200. Examples of such data include instructions for any application or method operating on the apparatus 200. The Memory 204 is implemented by at least any type of volatile or non-volatile Memory device or combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic Memory, flash Memory, magnetic disk or optical disk. Also stored in memory 204 are one or more modules configured to be executed by the one or more processors 218 to perform all or a portion of the steps of any of the methods described below in fig. 3-9.
The power supply component 206 provides power to the various components of the device 200. The power components 206 include at least a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for the device 200.
The multimedia component 208 includes a screen that provides an output interface between the device 200 and the user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a touch panel. If the screen includes a touch panel, the screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. The screen further includes an Organic Light Emitting Display (OLED for short).
The audio component 210 is configured to output and/or input audio signals. For example, the audio component 210 includes a Microphone (MIC) configured to receive external audio signals when the device 200 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals may further be stored in the memory 204 or transmitted via the communication component 216. In some embodiments, audio component 210 also includes a speaker for outputting audio signals.
The sensor component 214 includes one or more sensors for providing various aspects of status assessment for the device 200. For example, the sensor assembly 214 detects the open/closed status of the device 200, the relative positioning of the components, the sensor assembly 214 also detects a change in position of the device 200 or a component of the device 200, and a change in temperature of the device 200. In some embodiments, the sensor assembly 214 also includes a magnetic sensor, a pressure sensor, or a temperature sensor.
The communication component 216 is configured to facilitate wired or wireless communication between the apparatus 200 and other devices. The device 200 accesses a WIreless network based on a communication standard, such as WiFi (WIreless-Fidelity). In an exemplary embodiment, the communication component 216 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the Communication component 216 further includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wideband (UWB) technology, bluetooth technology, and other technologies.
In an exemplary embodiment, the apparatus 200 is implemented by one or more Application Specific Integrated Circuits (ASICs), digital signal processors, digital signal processing devices, programmable logic devices, field programmable gate arrays, controllers, microcontrollers, microprocessors or other electronic components for performing the methods described below.
FIG. 3 is a flow diagram illustrating a method for implementing operations in the construction of an artificial intelligence application, according to an example embodiment. In an exemplary embodiment, the method for implementing operation in the building of the artificial intelligence application, as shown in fig. 3, includes at least the following steps.
In step 310, an artificial intelligence algorithm configuration selection made by a user for the constructed artificial intelligence application is received, the artificial intelligence algorithm configuration being a network topology describing mathematical operations performed in the artificial intelligence application.
The artificial intelligence algorithm configuration is an algorithm logic description of a model and data adopted in the constructed artificial intelligence application, namely an algorithm logic formed by introduced mathematical primitives, and as indicated by the foregoing description, the artificial intelligence algorithm configuration is a network topology formed by a plurality of chunks.
On one hand, after the chunks in the graphical interface building area are added and linked along with the operation of the user on the chunks, the chunks distributed in the building area and the relationship among the chunks form a chunk framework for artificial intelligence application, namely an artificial intelligence algorithm configuration.
It should be noted that a set of blocks is a graphical representation corresponding to a set pattern or data, and a set of blocks is a graphical representation of a set model or data. The set model can be a single mathematical operation or a model implemented by integrating two or more mathematical operations, and is a complex formed by integrating input, output and mathematical operations, which is an independent unit, also called a mathematical primitive, no matter the single mathematical operation or the integration of two or more mathematical operations. The mathematical primitives correspond to chunks, which are arranged over the operator interface by chunks according to the algorithms that need to be involved in the built artificial intelligence application.
Mathematical primitives corresponding to the chunks define mathematical operations on one level. Of course, at a more subdivided level, the mathematical operations defined by the mathematical primitives will be subdivided, with the blocks corresponding to several mathematical operations, depending on the level. The mathematical primitives will define the chunks from the inputs, outputs, and mathematical operations performed, as well as themselves. In the building of the artificial intelligence application, along with the execution of the block configuration, the configuration of mathematical primitives related to the artificial intelligence application is carried out, so as to deploy the mathematical operations executed in the built artificial intelligence application, and control the input and the output of each mathematical operation for the execution of the mathematical operations.
On the other hand, the selection of the artificial intelligence algorithm configuration of the constructed artificial intelligence application by the user is the selection of the constructed artificial intelligence application, and the selection of the constructed artificial intelligence application is also the selection of the artificial intelligence algorithm configuration.
The selection of the artificial intelligence algorithm configuration is a process that a user selects the artificial intelligence application which needs to be operated currently, and for a plurality of built artificial intelligence applications, the user selects a certain artificial intelligence algorithm configuration which is suitable for the current data processing requirement according to the corresponding artificial intelligence algorithm configuration, so that the selection of the corresponding artificial intelligence application is realized, and the selected artificial intelligence application can be operated through the execution of the subsequent steps.
In an exemplary embodiment, this step 310 includes: and receiving and obtaining the artificial intelligence algorithm configuration selection of the user for the constructed artificial intelligence application according to the selection of the user for the constructed artificial intelligence application, wherein the artificial intelligence algorithm configuration selection is mapped with the dictionary stored for the artificial intelligence application.
As indicated by the foregoing description, the artificial intelligence algorithm configuration describes a logical algorithm for an artificial intelligence application that has been built. For example, it may be in the form of a graph, from which a graphical description of the logical algorithm is made.
For a built artificial intelligence application, the artificial intelligence application can be displayed through an artificial intelligence algorithm configuration, for example, an artificial intelligence algorithm configuration graph, so as to be selected by a user. The displayed artificial intelligence algorithm configuration can be obtained in artificial intelligence application construction performed by the user, can be shared by other users, can be preset by a system, and is not limited herein.
The artificial intelligence algorithm configurations corresponding to the built artificial intelligence applications are displayed in a list form, for example, a thumbnail list form by an artificial intelligence algorithm configuration diagram, and for a user, the user can select to operate the corresponding artificial intelligence application only by triggering and selecting any one artificial intelligence algorithm configuration.
For the constructed artificial intelligence application, artificial intelligence algorithm configuration and dictionary storage are carried out for the purpose, so that the user can call the artificial intelligence application at will. That is, each artificial intelligence algorithm configuration for which an artificial intelligence application has been built is mapped, i.e., a dictionary is stored in association.
In step 330, a dictionary describing the mathematical operations performed is selected based on the artificial intelligence algorithm configuration.
The artificial intelligence algorithm configuration is corresponding to the built artificial intelligence application, and for the artificial intelligence application, a dictionary describing executed mathematical operation is configured under the control of the corresponding artificial intelligence algorithm configuration. The dictionary is a structured data existing form of the artificial intelligence algorithm configuration, and the constructed artificial intelligence algorithm configuration is converted from the contained chunks to the dictionary along with construction of the artificial intelligence application, so that the dictionary formed by the mathematical elements corresponding to the chunks and the core parameters is obtained, and the core parameters are configured corresponding to the chunks.
Thus, with the selection of artificial intelligence algorithm configurations made, a dictionary is obtained for the selected artificial intelligence algorithm configuration, and the corresponding selected operational artificial intelligence application.
In one exemplary embodiment, the dictionary is stored in a server from which it is retrieved by a user triggering an artificial intelligence algorithm configuration selection at the front end.
Of course, the method is not limited to this, and the dictionary may be dynamically converted according to the artificial intelligence algorithm configuration selected by the user or even the dynamic adjustment performed on the artificial intelligence algorithm configuration, so as to obtain the dictionary suitable for the current user operation intention. In this exemplary embodiment, it can be understood that the artificial intelligence algorithm configuration available for the user to select will exist as a template, and the user only needs to perform dynamic adjustment on the basis to obtain the actually required artificial intelligence algorithm configuration, and then performs the conversion from the included chunks to the dictionary according to the artificial intelligence algorithm configuration. For the user, the artificial intelligence application is dynamically realized.
The artificial intelligence algorithm configuration indicates chunks included in the network topology of the built artificial intelligence application and the link relation among the chunks, and the chunks correspond to the mathematical primitives, so that the conversion of the chunks into the dictionary according to the artificial intelligence algorithm configuration is the conversion of each chunk into dictionary data, and the obtained dictionary data is used for forming the dictionary for realizing the built artificial intelligence application.
That is, for the artificial intelligence algorithm configuration, dictionary data is generated according to the contained chunks, so that the conversion from the chunks contained in the artificial intelligence algorithm configuration to the dictionary is realized. Dictionary data generated for the block is used to indicate the corresponding mathematical primitive, and input and output control is performed on this mathematical primitive.
Optionally, the configuration of the artificial intelligence algorithm is oriented to a network topology formed by the chunks, the mathematical primitive identifications and the core parameters configured for the mathematical primitives are obtained according to the mathematical primitives corresponding to the chunks, the mathematical primitive identifications are used as index items, the core parameters are used as index values to construct dictionary data corresponding to the chunks, and by analogy, the dictionary data corresponding to all the chunks form a built artificial intelligence application dictionary.
It should be added here that the core parameters are identified in correspondence with the mathematical primitives. That is, the core parameters are configured for the corresponding mathematical primitive.
In an exemplary embodiment, for many algorithms involved in artificial intelligence applications, the core parameters corresponding to the mathematical primitives include the key data for the hyper-parameters, input dimensions, and output dimensions for which the model is applicable. The configuration of the core parameters can ensure the smooth execution of the mathematical operation defined by the corresponding mathematical primitive.
In step 350, the mathematical operation is reconstructed on the dictionary through the mathematical language representation of the graph theory, and executable text for performing the mathematical operation is obtained.
And through the execution of the steps, the operation of the artificial intelligence application corresponding to the selected artificial intelligence algorithm configuration is initiated, the initiated and operated artificial intelligence application obtains a corresponding executable text through the dictionary, and the operation of the artificial intelligence application is further realized through the execution of the executable text.
In an exemplary specific implementation, after a dictionary is obtained for a currently selected artificial intelligence algorithm configuration, mathematical primitive identifications and core parameters related in the selected running artificial intelligence application can be obtained through the dictionary, on the basis, each mathematical primitive is continuously represented by a mathematical language of graph theory through a DAG data structure, a link relation between the mathematical primitive and other mathematical primitives is constructed, and then an executable text of the selected running artificial intelligence application is finally obtained. The executable text contains all executable sentences of the constructed artificial intelligence application and is composed of code information of all mathematical primitives.
Each mathematical primitive has a corresponding code description, i.e., code information including core parameters, for implementing the execution of the corresponding mathematical operation, so as to implement the execution of the corresponding mathematical operation by the execution of the corresponding code description. The artificial intelligence algorithm configuration formed by the chunks, and the existence of the corresponding mathematical primitive indicates the code description information for executing a series of mathematical operations under the artificial intelligence algorithm configuration, namely, the executable text formed by the code information which is corresponding to the mathematical primitive identification and contains the core parameter.
In step 370, the user is enabled to run the built artificial intelligence application according to the selected artificial intelligence algorithm configuration through execution of the executable text.
After the executable text of the artificial intelligence application is obtained, the running of the artificial intelligence application can be triggered, and at the moment, the deployed function of the artificial intelligence application is realized through the execution of the statements in the executable text, so that the artificial intelligence requirement of a user is met.
The artificial intelligence algorithm configuration, in many cases in the form of a graph, i.e. it may be in the form of an artificial intelligence algorithm configuration graph. According to the method and the device, the algorithm diagram, namely the translation of the artificial intelligence algorithm configuration, is realized, so that the artificial intelligence application is accurately realized, and therefore, for a user, only approximate algorithm logic construction meeting the personalized requirements of the artificial intelligence application of the user is needed, actual programming is not needed, and graphical programming can be realized by facing to the artificial intelligence application technology, so that the required artificial intelligence application is obtained.
By the implementation of the exemplary embodiment, the gradually emerging graphical programming can be used for the implementation of artificial intelligence applications, and is no longer better than the building of simple algorithm logic, for example, is no longer limited to programming implementation oriented to childhood education, and the implementation of graphical programming not only implements simple applications, but also implements artificial intelligence applications with complex algorithms.
The artificial intelligence application can be built through chunk splicing, namely after the chunks are selected and spliced through dragging the chunks, the spliced chunks are translated into the artificial intelligence application through the exemplary embodiment, and the difficulty and the bottleneck of programming related to the artificial intelligence application through graphical programming are solved.
In the prior art, graphical programming of artificial intelligence application is difficult to achieve by building blocks by means of the chunks, and artificial intelligence application building facing graphical programming can be achieved under the exemplary action.
In another exemplary embodiment, this step 310 includes: and according to the artificial intelligence application construction selected by the user, receiving a character string corresponding to the dictionary, wherein the dictionary corresponding to the character string is used for recording the artificial intelligence algorithm configuration formed by the artificial intelligence application constructed by the user through mathematical primitive configuration at present.
Correspondingly, step 330 includes: and decoding the character strings and converting the character strings into a dictionary to obtain the dictionary formed by taking the mathematical primitive identifications and the core parameters as dictionary data, wherein the mathematical primitive identifications and the core parameters in the dictionary data correspond to the mathematical primitives for building artificial intelligence application configuration.
This exemplary embodiment, among other things, corresponds to the artificial intelligence application building process currently being performed by the user. With the construction of the artificial intelligence application of the user, a character string corresponding to the dictionary is obtained, and the dictionary corresponding to the character string is a data existence form of the artificial intelligence algorithm configuration corresponding to the artificial intelligence application currently constructed by the user. The artificial intelligence algorithm configuration describes the algorithm logic related to the artificial intelligence application built by the user, and the algorithm logic is visually described in the form of a graph, namely an artificial intelligence algorithm configuration graph, on one hand, and is recorded in the form of a dictionary through structured data on the other hand.
The user conducts artificial intelligence application construction on the front end through control applied to the chunks on the graphical interface, so that artificial intelligence algorithm configuration is obtained, the artificial intelligence algorithm configuration is converted into a dictionary, character string conversion is conducted on mathematical primitive identification and index core parameters in the dictionary, and then character strings are transmitted to the server for the artificial intelligence constructed by the user.
This is the implementation performed in the front end, i.e. the user terminal used by the user. And the back end, namely the server, is suitable for artificial intelligence application building selected by a user, receives and obtains the character strings corresponding to the dictionary, decodes and converts the character strings into the form of the dictionary, and then decodes through the mathematical language representation of the graph theory to obtain the executable text.
In an exemplary embodiment, for the artificial intelligence application building performed by the user at the front end, the artificial intelligence application building can be implemented at the user terminal through the following steps:
(1) receiving a user selection instruction for a chunk on a graphical interface, the chunk being a graphical representation of a corresponding mathematical primitive;
(2) setting up and configuring chunks for artificial intelligence application in a setting-up area of a graphical interface by selecting an instruction, and mutually linking the chunks to form an artificial intelligence algorithm configuration of the artificial intelligence application set up under the control of a user;
(3) and converting the contained chunks into a dictionary according to the artificial intelligence algorithm configuration to obtain the dictionary formed by the mathematical primitive identifications corresponding to the chunks and the core parameters, wherein the core parameters are configured corresponding to the chunks.
The graphical interface is a user interface for freely establishing artificial intelligence application, the indicated artificial intelligence application is established, namely the configuration of the related sequential execution mathematical operation is realized, the configured and executed mathematical operation forms the algorithm realization of the artificial intelligence application, and the artificial intelligence application meeting the set requirement is realized.
Optionally, the graphical interface includes an operation component selection bar, an operation interface, a toolbar, and an operation result display area. The selection command applied to the chunk is triggered in the operating element selection field. The operating component selection bar includes a plurality of selectable chunks corresponding to the initially configured model and/or data. The user can trigger the operation on the chunk to generate a selection instruction on the chunk, so as to complete the selection of the corresponding chunk, and at this time, the selected chunk is configured on the operation interface.
For the operation component selection column, on one hand, the mathematical primitives involved in the selection of the artificial intelligence application currently built are used, for example, the mathematical primitives may be models integrated together, and on the other hand, data used by iterative training may also be selected for the artificial intelligence application built, and whatever selection is selected, the data used in iterative training will exist in the form of chunks on the operation component selection column of the graphical interface.
The operation interface is a building area on the graphical interface and used for placing the selected chunks for building of artificial intelligence application, and the chunks are placed under the control of a user so as to realize building. After the chunk in the selection column of the operation component is triggered, the operation component is placed in the operation interface. For example, in the chunk selection, the chunk in the operation component selection column can be dragged to the operation interface through the dragging operation applied to the chunk, and the chunk dragged to the operation interface is used for building the artificial intelligence application.
The tool bar on the graphical interface is used for adjusting the corresponding data and model for the configured chunks in the construction of the artificial intelligence application, for example, fine-tuning the data and model.
And the operation result display area on the graphical interface is used for displaying the result after the constructed artificial intelligence application is operated. For example, after iterative training is performed on the currently-built artificial intelligence application, the corresponding iterative training situation and the obtained classification effect are displayed in the operation result display area.
In summary, in the graphical interface displayed on the user terminal for the construction of the artificial intelligence application, the chunks configured in the construction area are used for the construction of the artificial intelligence application by configuring the chunks provided by initialization in the construction area.
Therefore, the chunks configured in the building region, namely the chunks selected for the currently built artificial intelligence application, can be determined, the mathematical primitives corresponding to the selected chunks are used for forming the currently built artificial intelligence application, and for the building of the artificial intelligence application, the what you see is what you get artificial intelligence application is realized, and both the degree of freedom and the adaptivity of the built artificial intelligence application can be enhanced.
In addition, the dragging operation, the selecting operation and the like applied to the chunks needing to be selected are realized in the building area, namely, the chunks selected by the user are configured in the building area along with the selection instruction of the chunks on the graphical interface.
The selection instruction indicates the chunk selected by the user. As mentioned above, a chunk is a graphical representation of a corresponding mathematical primitive, so that the chunk can be uniquely identified by the corresponding mathematical primitive identifier, and the selection instruction carries the mathematical primitive identifier to identify the chunk selected by the user and execute a corresponding response thereto.
And when the chunk selection is performed once, the process for realizing the artificial intelligence application construction receives a selection instruction of the corresponding chunk on the graphical interface. By analogy, as chunk selection continues, selection instructions corresponding to different chunks will be received continuously.
In an exemplary embodiment, the step of receiving a user selection instruction for a chunk on the graphical interface includes: in a plurality of chunks initially configured on the graphical interface, a selection instruction of the chunks on the graphical interface is received through user operation applied to the chunks until all chunks for realizing artificial intelligence application are selected.
Wherein the graphical interface is initially configured with a plurality of chunks, e.g., the existence of a chunk in the operating component selection bar, as previously described. The user can continuously apply user operation on the required chunks according to the building requirement of the artificial intelligence application, for example, drag operation to the building area, so as to generate a selection instruction for the continuously selected chunks, and the corresponding process continuously receives the generated selection instruction until all the required chunks are selected.
Along with the receiving of the selection instruction, a configuration chunk is continuously built for artificial intelligence application in a building area of the graphical interface, and the configured chunk is corresponding to the mathematical primitive identification carried in the selection instruction.
Under the action of the selected instruction, the building area of the graphical interface configures the chunks corresponding to the related mathematical primitives for the artificial intelligence application, so that more than two chunks exist in the building area of the graphical interface.
It should be appreciated that as chunk selection proceeds on the graphical interface, the selected chunks are scattered across the build area of the graphical interface. At the moment, the mutual linkage between the chunks is carried out under the control of the user, so that the artificial intelligence algorithm configuration of the constructed artificial intelligence application is obtained.
An artificial intelligence algorithm configuration is a network topology that describes the mathematical operations performed in an artificial intelligence application, in other words, the network topology formed by deployed mathematical primitives. For the interlinked mathematical primitives, the output of the previous mathematical primitive is used as the input of the next mathematical primitive, and so on to form the whole artificial intelligence algorithm configuration.
In one exemplary embodiment, the artificial intelligence algorithm configuration is in the form of a graph, i.e., it may be in the form of an artificial intelligence algorithm configuration graph. The artificial intelligence algorithm configuration indicates the chunks included in the constructed network topology and the link relationships between the included chunks.
On a graphical interface for artificial intelligence application building, as the user operates the set artificial intelligence application to freely select and configure the chunks and build the link relation between the selected configuration chunks, the chunks are freely operated by the user according to the own artificial intelligence application requirements, graphical programming of the artificial intelligence application building is realized, and the threshold is reduced to the maximum extent.
In addition, the core parameters recorded for each chunk by the dictionary are obtained for the chunk in the building area. That is, the core parameters configured to correspond to the mathematical primitives are obtained through data and model fine-tuning performed in the toolbar by the corresponding components in the build area.
At the front end, decoding of artificial intelligence application built by a user can be initiated to the back end through the obtained dictionary, and the artificial intelligence application is triggered to run at the back end.
After the dictionary is obtained by converting the current constructed artificial intelligence application, the decoding of the constructed artificial intelligence application can be initiated to the server through the dictionary, so that the operation of the artificial intelligence application at the server is triggered.
The dictionary is obtained by building an artificial intelligence application by a user, and the built artificial intelligence application can be obtained by the server side in a dictionary form through the identification of the mathematical primitives and the core parameters of the contained dictionary data records.
Through the exemplary embodiment, the artificial intelligence application building based on the graphical interface is carried out on the user side realized by the user terminal, namely, graphical programming is carried out, after the user completes the building of the artificial intelligence application, the built artificial intelligence application is generated by facing to the mathematical primitives corresponding to the chunks, so that the artificial intelligence application building carried out by the user can be known by the server, and then the artificial intelligence application building runs on the server, thereby meeting the artificial intelligence requirements of the user and obtaining the functions which can be provided by the built artificial intelligence application.
For the construction of the artificial intelligence application, under the action of a graphical interface and the chunks corresponding to the mathematical primitives, the development of the complex algorithm related to the artificial intelligence application is realized for users, namely, the required complex algorithm is constructed by selecting and mutually linking different chunks, but the users do not need to have code development capability and learn the professional knowledge of the artificial intelligence, only the functions of the mathematical primitives corresponding to each chunk need to be known approximately, and the threshold of the construction of the artificial intelligence application is eliminated for the public.
The above-mentioned exemplary embodiment packages the algorithm involved in the artificial intelligence application, i.e. the mathematical operation and the input and output thereof, as the mathematical primitive, and presents the mathematical primitive to the user in the form of chunks, so that the construction of the artificial intelligence application on the graphical interface is performed on the basis of the mathematical primitive.
On the basis, for the execution of the step of constructing and configuring the chunks for the artificial intelligence application in the construction area of the graphical interface by selecting the instructions and mutually linking the chunks to form the artificial intelligence algorithm configuration of the constructed artificial intelligence application under the control of the user, in an exemplary embodiment, the method comprises the following steps:
(1) configuring a chunk indicated by a selection instruction in a building area of a graphical interface;
(2) acquiring corresponding core parameters for the chunks placed in the building area;
(3) with the configuration of more than two chunks in the building area, the chunks are mutually linked to form an artificial intelligence algorithm configuration of the artificial intelligence application under the control of a user.
And carrying out chunk configuration on the building area according to the received selection instruction along with the reception of the selection instruction, so that the chunk indicated by the selection instruction is added to the building area of the graphical interface.
Here, the step (1) is executed in a process of adding required mathematical primitives step by step for the currently built artificial intelligence application, so as to finally construct an artificial intelligence algorithm configuration for realizing the artificial intelligence application, that is, to realize the network topology of the mathematical primitives corresponding to the artificial intelligence application.
Along with the operation of the user, the selection instruction is triggered and generated continuously, and then the chunk selected by the user is added in the building area continuously. And for each chunk, the configuration and adjustment of the corresponding core parameters can be carried out so as to be suitable for the artificial intelligence application which is currently built.
In an exemplary embodiment, for a set of blocks placed in a building area, configuration and adjustment of core parameters in corresponding mathematical primitives are performed, and after the configuration and adjustment of the core parameters of the set of blocks are completed, configuration and adjustment of the core parameters of other sets of blocks can be performed.
For example, core parameter configuration and adjustment for a chunk in a build area may be initiated by selection of the chunk. Specifically, after a group of blocks in the building area is selected, a toolbar on the graphical interface is used for configuring and adjusting core parameters of the group of blocks, and at this time, a user only needs to configure and adjust set parameters on the toolbar.
The core parameters include the hyper-parameters used by the model in the mathematical primitive, as well as the input dimensions, the output dimensions. The link relation transformation of the corresponding chunks not only needs to transform other chunks linked by the chunks in the building area, but also adjusts the input dimension and/or the output dimension in accordance with the mathematical primitives of the linked chunks so as to adapt to the building of the dynamically-changing artificial intelligence application.
In one exemplary embodiment, for step (2), it comprises: and building a selected chunk for the graphical interface, and acquiring a core parameter corresponding to the chunk according to the core parameter configuration of the user on the chunk.
As indicated in the foregoing description, at least one chunk is distributed in the construction area, and for any chunk, the corresponding core parameter configuration process can be initiated through its selection on the construction area. Once a set of blocks is selected at the build area, core parameter configuration of the set of blocks can be performed.
The core parameter configuration is the process of inputting parameters, adjusting parameters, selecting parameters and the like under the control of a user, and the different mathematical elements corresponding to different blocks correspond to different core parameter configuration processes.
In any case, the core parameters corresponding to the selected chunks in the building area are obtained through the core parameter configuration, and the core parameters corresponding to all chunks in the building area are obtained by analogy, so that the core parameter configuration of all chunks in the building area is completed, and the operation of the built artificial intelligence application is further ensured.
For more than two chunks stored in the building area, the chunks are linked under the control of a user, so that the chunks distributed in the building area can be built into an artificial intelligence algorithm configuration applied to artificial intelligence.
The linkage between the chunks refers to a connection line of the chunks in the building area, and the connection line indicates the input and output relation between the connected chunks besides the mutual connection between the chunks; on the other hand, the link between the chunks also indicates that the output of the previous chunk will be the input of the linked next chunk.
By analogy, an artificial intelligence algorithm configuration is formed along with the end of the added chunks in the building area and the setting of the links among the chunks. The artificial intelligence algorithm configuration is an algorithm description of an artificial intelligence application currently set up by a user. For the artificial intelligence application currently set up by the user, the running process of the artificial intelligence application is a process of executing the algorithm according to the mathematical primitives indicated by the chunks in the corresponding artificial intelligence algorithm configuration and the mutual linking relation.
In one exemplary embodiment, the inter-linking between the chunks is achieved by the connection between the chunks performed under the user operation. The obtained artificial intelligence algorithm configuration is realized under the control of a user, the control of the user is the operation of selecting the chunks and the operation of connecting the chunks, which are required to be triggered by the user when the artificial intelligence application is built, and the like, and the operation of randomly configuring the chunks for the built artificial intelligence application and constructing the links among the configured chunks is the user operation triggered by the formation of the artificial intelligence algorithm configuration.
According to the method, the artificial intelligence algorithm configuration is built for the currently built artificial intelligence application, algorithm development is achieved, the artificial intelligence algorithm configuration is randomly built for the user's own needs by the user, flexibility is enhanced, meanwhile, the needed artificial intelligence application building is achieved, and the method is not limited by the loss of professional knowledge such as code knowledge, artificial intelligence algorithm and mathematical representation.
For the step of converting the contained chunks into a dictionary according to the artificial intelligence algorithm configuration, and obtaining the dictionary formed by the mathematical primitive identifications corresponding to the chunks and the core parameters, in an exemplary embodiment, the method includes:
acquiring mathematical primitive identifications and core parameters corresponding to the chunks for the chunks contained in the artificial intelligent algorithm configuration;
and constructing a dictionary of the artificial intelligence application by taking the mathematical primitive identifications as index items and the core parameters as index values.
And correspondingly completing the construction of the artificial intelligence algorithm configuration for the construction area in which the chunk addition and the link are completed. Under the artificial intelligence algorithm configuration, a dictionary is generated for the built artificial intelligence application by acquiring the mathematical primitive identification and the core parameters corresponding to each group of blocks.
It should be understood that the generated dictionary is used for transmitting the artificial intelligence algorithm configuration which is built by the user in an adaptive mode to the server side, and therefore the artificial intelligence application which runs on the server side is obtained.
Each chunk has a corresponding mathematical primitive, that is, the chunk is a graphical representation of the corresponding mathematical primitive, so that for the chunk under the artificial intelligence algorithm framework, the corresponding mathematical primitive identification can be obtained, and the core parameters can be obtained along with the core parameter configuration on the chunk. For a set of blocks, the obtained core parameters are identified corresponding to the mathematical primitives.
And constructing dictionary data for each chunk under the artificial intelligence algorithm configuration, namely constructing dictionary data of the chunk by taking the corresponding mathematical primitive identification as an index item and taking the core parameter as an index value, and so on, wherein the dictionary data of all chunks under the artificial intelligence algorithm configuration form a dictionary for artificial intelligence application.
Under the action of the dictionary, the image information of the chunk is converted into codes existing in a server, namely executable texts of artificial intelligence application are obtained, so that the dilemma that an artificial intelligence algorithm is complex and is difficult to adapt to user development is solved.
In addition, by means of dictionary generation, core parameters are transmitted for the artificial intelligence application built by the user, so that the core parameters of user personalized configuration are provided for the server, and the artificial intelligence application can be accurately adapted to the artificial intelligence application requirements of the user when built.
For the back end implemented by the present invention, the acquisition of the character string corresponding to the dictionary is implemented under the coordination of the character string conversion and transmission performed by the front end. Therefore, for step 310, before this, the user terminal will perform string conversion on the mathematical primitive identifiers and the indexed core parameters in the dictionary, then transmit the string to the back-end for the artificial intelligence application constructed by the user, and initiate decoding of the string by the back-end through transmission of the string to obtain the executable text of the artificial intelligence application to run on the server.
After the dictionary is generated for the artificial intelligence application built for the user through the foregoing exemplary embodiment, the character string conversion needs to be performed on the dictionary, so that the dictionary carries the mathematical primitive identification and the core parameters, and the transmission from the user terminal to the server is facilitated.
In an exemplary embodiment, the mathematical primitive identifications in the dictionary and the core parameters of the index are converted into JSON character strings, and the dictionary is transmitted to the server side in the form of the JSON character strings.
Executable text is a code description of an artificial intelligence application built by a user. The execution of the executable text executes a series of operations configured by the user for realizing the artificial intelligence application, and the process is the running process of the artificial intelligence application.
Through the exemplary embodiment, the constructed artificial intelligence application can exist in the server, and further, the configured functions in the artificial intelligence application can be fully realized through excellent hardware conditions of the server.
For the user, on the one hand, the artificial intelligence application is built rapidly and freely through the graphical interface, and on the other hand, the built artificial intelligence application exists and runs at the rear end, namely the server, so that the built artificial intelligence application can obtain excellent hardware performance and strong computing power, and the performance of the built artificial intelligence application is enhanced.
Fig. 4 is a flowchart illustrating a description of step 350 according to a corresponding embodiment of fig. 3. In an exemplary embodiment, this step 350, as shown in FIG. 4, includes at least the following steps.
In step 351, the data missing code of the corresponding mathematical operation is obtained according to the mathematical primitive identification in the dictionary.
In step 353, the core parameters corresponding to the mathematical primitive identifier are filled in the obtained data missing code, so as to obtain the code information for executing the corresponding mathematical operation.
In step 355, the executable text of the artificial intelligence application is reconstructed from output to input sequentially through the mathematical language representation of graph theory according to the input dimension and the output dimension indicated in the core parameter corresponding to the mathematical primitive identification.
The realization of the process can reconstruct and execute code information of corresponding mathematical operation through the mathematical primitive identification and the core parameter contained in the dictionary to obtain the executable text of artificial intelligence.
After the converted character string is decoded to obtain the dictionary, a set of mathematical primitive identifications and core parameters corresponding to each mathematical primitive can be obtained from the obtained dictionary.
As mentioned above, the dictionary data of the mathematical primitive identification and the core parameter indicates the corresponding mathematical primitive and other mathematical primitives linked to the mathematical primitive, i.e. it is additionally agreed that the mathematical primitive whose dimension is the output dimension of the core parameter is linked to the current mathematical primitive.
Therefore, the reconstruction of the mathematical primitive can be performed through a piece of dictionary data contained in the dictionary, the reconstructed mathematical primitive is represented in the form of code information in the aspect of program execution, and the code information is obtained through the reconstruction, wherein the code information is the program language for executing the mathematical operation corresponding to the mathematical primitive. By analogy, all code information constitutes executable text for artificial intelligence applications.
Specifically, the back end is controlled by artificial intelligence application construction performed by the user terminal to obtain a dictionary for the constructed artificial intelligence application, and the dictionary records mathematical primitive identifications and core parameters corresponding to each mathematical primitive used by the constructed artificial intelligence application.
In addition, the back-end stores information about the code for the mathematical operations that need to be performed. The code related information stored for the execution of each mathematical operation is data-missing code that has missing core parameters. The data missing codes are correspondingly stored by taking the mathematical primitive identifications as indexes. The data missing codes of the core parameters are missing, different core parameters can be filled in a suitable way along with the realization of different artificial intelligence applications, and different configurations of executed mathematical operations are realized, so that the realization of the artificial intelligence applications is fully adapted.
Under the action of the data missing codes stored at the back end, the back end provides an artificial intelligence application algorithm for the user under the support of the data missing codes, so that the user can simply and intuitively build the artificial intelligence application through the chunks, and on the other hand, the algorithm implementation which is flexible and free is obtained under the support of the data missing codes, and the flexibility of building the artificial intelligence application is enhanced.
As mentioned above, the core parameters in the dictionary are corresponding to the mathematical primitive identifications, so that after the data missing code is obtained from the mathematical primitive identifications, the corresponding core parameters are also obtained from the mathematical primitive identifications, and the obtained core parameters are filled into the data missing code to obtain the complete code information for executing the corresponding mathematical operation.
The code information obtained by filling the core parameters is the mathematical primitive description on the code level. Under the action of data missing codes and code information, a user can build an artificial intelligence application according to the requirement of the user even if the user does not have programming intelligence and programming skills.
As indicated in the foregoing description, the core parameters indicate the input dimension and the output dimension for the corresponding executed mathematical operation and for the implementation of the artificial intelligence algorithm by the code information, and therefore can be interfaced with other mathematical primitives based thereon.
It should be noted here that the mathematical language representation of the graph theory is directed to the mathematical primitives and also to the code information for performing the corresponding mathematical operations. The mathematical language representation of the graph theory is to rapidly and accurately reconstruct the input, output and mathematical operation corresponding to the mathematical primitive by means of the DAG structure, and then continuously reconstruct the DAG structure corresponding to the mathematical primitive according to the sequence from the output to the input on the basis until the reconstruction of all the mathematical primitives corresponding to the dictionary is completed.
After all mathematical primitives corresponding to the dictionary are reconstructed to obtain the mathematical language representation of the graph theory, splicing the corresponding code information according to the mathematical language representation of the graph theory to obtain the executable text of the artificial intelligence application.
For example, FIG. 5 is a diagram illustrating a DAG structure corresponding to a simple mathematical primitive in accordance with an illustrative embodiment. The simple mathematical primitive is "I + W ═ O", i.e. with "I" as input and "O" as output, the addition operation is performed on the input data, and "W" is the parameter obtained by the required training.
The following table is an "I" and "O" data table configured by a user on a chunk Add in a graphical interface, where the chunk Add indicates that a corresponding mathematical primitive is to perform an addition operation on input data, and specifically is as shown in the following table:
| I | O |
| 1 | 7 |
| 2 | 8 |
| 3 | 9 |
| 4 | 10 |
| 5 | 11 |
TABLE 1
With this exemplary embodiment, building of artificial intelligence applications by means of DAG structures can be quickly achieved even if complex artificial intelligence algorithms are involved.
The operation of the artificial intelligence application set up by the user in the server is performed by means of the received character strings, namely, the server receives the character strings corresponding to the dictionaries in the artificial intelligence application set up selected by the user, and the dictionaries corresponding to the character strings are used for describing chunks configured for the artificial intelligence application set up.
In order to realize the construction of the artificial intelligence application in the user terminal, the deployed server responds to the operation and control of the user terminal to cooperate with the user terminal to realize the construction of the artificial intelligence application, and the artificial intelligence application existing in the server is obtained for the user.
The operation of the artificial intelligence application is realized by executing a series of mathematical operations, namely, the arithmetic required by a user is realized by taking a mathematical primitive as a unit. The execution of the mathematical operation is necessarily supported by the corresponding code information, so that the execution of the mathematical operation by the server is controlled. Therefore, the server stores the information related to the codes for the deployed chunks, and the users need to respectively call the chunks under the control of the dictionary generated by the built artificial intelligence application, so as to obtain the executable text capable of realizing the running of the artificial intelligence application.
Based on the method, as the artificial intelligence application built by the user at the user terminal is completed, the dictionary generated for the built artificial intelligence application is converted into the character string and is sent to the server, so that the server can obtain the artificial intelligence application built by the user. At this time, the server receives the character string sent by the user terminal. The dictionary converted into this string describes and defines the corresponding mathematical primitives for the respective chunk by a strip of dictionary data.
It should be understood that the mathematical primitives are operation units for implementing artificial intelligence application in the artificial intelligence application building implemented facing the user, and the operation units can be divided in different layers according to the needs, and then the mathematical primitives and corresponding chunks are configured for the operation units.
For example, the neural network operation, the point multiplication operation, the matrix multiplication operation, and the like can be configured correspondingly as operation units according to an artificial intelligence algorithm, and some neural network operations include the point multiplication operation, so that it can be seen that the two operations are operation unit divisions on different levels, but do not affect the configuration of the corresponding mathematical primitives and chunks.
Therefore, the subdivided operation units can be combined together to form a new operation unit on the previous level, namely, a mathematical primitive and a chunk, which is the chunk adding process referred to in the following exemplary embodiments.
Fig. 6 is a flowchart illustrating a description of step 370, according to a corresponding embodiment of fig. 3. In an exemplary embodiment, as shown in FIG. 6, this step 370 includes:
in step 371a, sample data suitable for artificial intelligence applications is obtained in a training mode, which is an operational mode with the executed artificial intelligence application.
In step 373a, the constructed artificial intelligence application is run through execution of the executable text, and parameter training is performed through sample data in the running of the artificial intelligence application.
The training process of the artificial intelligence application is provided for the artificial intelligence application built by the user, iterative training of parameters is completed for the artificial intelligence application built by the user through the set sample parameters, and the built artificial intelligence application can be normally used.
Sample data suitable for artificial intelligence application is configured by a user in an exemplary embodiment, for example, the user imports the sample data through a toolbar set by a graphical interface, and in addition, the set of the sample data can be realized by operating a chunk related to the data on the interface to set up the artificial intelligence application, and the selection of the sample data can be realized through a marquee set in the toolbar.
That is, the sample data is collected by the user, may be stored in the server and shared by other users, or may be preconfigured by the server according to the building requirements of various artificial intelligence applications, which is not limited herein.
After the construction of the artificial intelligence application is completed, namely all the chunks of the artificial intelligence application are added and completed in the construction area and are linked with each other, and then a training mode can be selected to perform parameter iterative training on the constructed artificial intelligence application.
That is to say, for the artificial intelligence application which is built, the training mode can be set through the user operation, and the operation decision mode can also be set, so that the operation of the artificial intelligence application is controlled.
For example, a switch which is turned on in a training mode or a decision mode can be arranged on the graphical interface, and a user can select to enter the corresponding mode only by operating the switch.
In the training mode, sample data suitable for artificial intelligence application is acquired according to sample data configuration performed by a user, for example, sample data recommended by a server is acquired, sample data shared by other users is acquired, and the required sample data is imported into a graphical interface by the user.
The artificial intelligence application is realized based on machine learning, and solves a specific problem for a user through an algorithm of the machine learning. For example, the machine learning algorithm may be a neural network algorithm. Therefore, the built artificial intelligence application necessarily involves the realization of training, and parameters required by the normal operation of the artificial intelligence application can be obtained through an iterative training party.
Through the exemplary embodiment, an entrance for acquiring the sample data is realized for the constructed artificial intelligence application, even if a user is limited by various conditions and cannot acquire the sample data with huge data volume, the sample data can be acquired by means of the server, and further training and subsequent use of the constructed artificial intelligence application are guaranteed.
In addition, the sample data submitted to the server by the user can be hidden according to self selection, namely stored as private data of the user, but the state of the sample data can also be set to be a public state so as to be shared in the server.
The server may obtain the sample data in a crowdsourcing manner, in addition to the sample data configured in advance and the sample data shared by the user, which is not limited herein.
Fig. 7 is a flowchart illustrating a description of step 370 in another exemplary embodiment according to a corresponding embodiment of fig. 3. In an exemplary embodiment, as shown in FIG. 7, this step 370 includes at least:
in step 371b, data selected for input by the user is obtained, the data including target data of the artificial intelligence application process set up by the user.
In step 373b, the trained artificial intelligence application is run in the decision mode to complete the processing of the user selected input data by executing the executable file that completes the parameter training.
According to the data processing method and the data processing device, the data decision of the user is achieved by using the built artificial intelligence application, and the built artificial intelligence application really meets the data processing requirement of the user.
The data selected and input by the user is the data which needs to be processed by the aid of the constructed artificial intelligence application, and it should be understood that the user is the artificial intelligence application construction which is performed for the processing requirement of the data.
Similar to the training mode, the processing of the user selection input data can be performed only by starting the decision mode under the user control.
In one exemplary embodiment, this step 371b includes: and receiving target data uploaded by the user for the constructed artificial intelligence application, wherein the target data is data for requesting the constructed artificial intelligence application to process.
That is to say, the user terminal receives the data uploaded to the artificial intelligence application built by the user on the graphical interface, determines the artificial intelligence application built by the user according to the user identification carried in the data, and triggers the determined artificial intelligence application to operate on target data in the data, which is requested to be processed by the artificial intelligence application built by the user.
After the artificial intelligence application is built and trained, a user uploads data required to be processed by the artificial intelligence application on a graphical interface, and the data content and the data type of the data are different according to different artificial intelligence applications.
The server is matched with a plurality of user terminals to realize the artificial intelligence application building on each user terminal, so that the artificial intelligence applications built by different users are different. Therefore, for the server, the dictionary is stored by taking the user identification as the index, so as to record the artificial intelligence applications corresponding to different users.
The built and trained artificial intelligence application is completed, and the dictionary of the artificial intelligence application is stored in the server so as to be called by the user at any time. That is, the user can invoke the built and trained artificial intelligence application as required along with accessing the server, and continuously iteratively optimize parameters in the artificial intelligence application by means of decision results in the use of the artificial intelligence application, so as to continuously improve the accuracy of the decision.
The user identification is used for uniquely marking the user, so that the user can log in the server through the user identification, and the constructed artificial intelligence application is called.
Through the execution of the steps, after the uploading of the data to be processed is completed, the artificial intelligence corresponding to the user identification is requested to process the uploaded data, and therefore the uploaded data is used as input to run the artificial intelligence application.
FIG. 8 is a flowchart illustrating a method for implementing operations in the construction of an artificial intelligence application, according to another illustrative embodiment. In another exemplary embodiment, the executable mathematical operation corresponds to a mathematical primitive, and the artificial intelligence algorithm configuration is a network topology formed by interlinked mathematical primitives, as shown in fig. 8, the implementation method of the operation in the construction of the artificial intelligence application at least comprises the following steps.
In step 510, receiving an update of a new mathematical primitive from a user, and obtaining mathematical primitive definition information, where the mathematical primitive definition information includes an identifier of the new mathematical primitive and at least one identifier of the mathematical primitive and a core parameter corresponding to the new mathematical primitive encapsulated under the identifier of the new mathematical primitive.
In step 530, the mathematical primitive definition information is saved, so that the graphical interface is initialized and configured with the newly added chunks.
The process is suitable for chunk adding performed by the user terminal, and is realized by adding chunks through user operation at the user side and further adding a server corresponding to the mathematical primitives.
At a user terminal, receiving a newly added chunk selection instruction of a user for chunks configured in a building area, wherein the newly added chunk selection instruction acts on at least one chunk configured in the building area; then generating a dictionary according to the chunk corresponding to the action of the newly-added chunk selection instruction, wherein the dictionary comprises mathematical primitive identifications and core parameters corresponding to the chunks; and (3) performing combined packaging on the mathematical primitive identifications and the core parameters in the dictionary into new mathematical primitives through mathematical language representation of graph theory, and finally updating the new mathematical primitives to the graphical interface and the server.
Therefore, the execution of steps 510 to 530 is the new mathematical primitive update performed by the server.
The building area is controlled by a user to add at least one chunk, and the process of adding the chunk to the building area through user control can be a building process of artificial intelligence application and a chunk newly-adding process for subsequent artificial intelligence application building.
In the construction of artificial intelligence application or under other conditions, the chunks can be newly added, so that the chunks initially configured on the graphical interface are newly added according to the requirements of the user, and the subsequent artificial intelligence application construction is facilitated.
In the artificial intelligence application building by users, there are situations where several chunks often need to be combined together, for which chunks can be combined together to build a new added chunk. The mathematical primitives corresponding to the newly added chunks are the combination of the mathematical primitives corresponding to these chunks.
In the building area, the new chunk adding can be initiated for the at least one selected chunk along with the selection of the at least one chunk by the user, and then a new chunk adding selection instruction is generated. The new chunk selection instruction is applied to at least one chunk selected by the user in the construction area.
And after receiving and obtaining a new chunk selection instruction generated by triggering of a user, generating a dictionary for the chunk acted on by the new chunk selection instruction.
Similar to building artificial intelligence applications, the implementation of newly added chunks still requires the generation of dictionaries for this purpose. And the dictionary generated for the newly added chunks performs corresponding mathematical primitive identification and core information description on the combined chunks. For the newly added blocks and the corresponding newly added mathematical primitives, the generated dictionary is described by data for this purpose.
In at least one chunk combined together to form the newly added chunk, in a dictionary generated for the newly added chunk, the mathematical primitive identifier uniquely identifies the corresponding chunk, and the core parameters corresponding to the chunk indicate the link relation between the chunks because the core parameters include the input dimension and the output dimension corresponding to the mathematical operation.
In particular, the core parameters specifying the input dimension and the output dimension for a chunk, i.e., a mathematical primitive, will be used to determine the links between chunks. The output dimension of a set of blocks of corresponding mathematical primitives indicates that the two blocks are interlinked if the output dimension is the input dimension of another set of blocks of corresponding mathematical primitives.
Therefore, the combined chunks and the link relation between the chunks can be accurately known through the generated dictionary without consuming excessive storage space and more calculation cost.
The dictionary is used as the data description of the new added mathematical primitive, and the data description in the dictionary is expressed as the mathematical language of the graph theory according to the mathematical primitive identification and the corresponding core parameter, so as to obtain the new added mathematical primitive formed by combining and packaging a plurality of groups together.
It should be noted that the mathematical language representation of the graph theory is to track the input values, the mathematical operations performed on the input values, and the output values in the basic mathematical primitive to reconstruct and characterize the mathematical primitive. Whether input values, mathematical operations or output values are reconstructed in the form of nodes, thereby forming a mathematical linguistic representation of the mathematical primitives on a graph theory. The newly added mathematical primitives can be accurately and quickly reconstructed through the mathematical language representation of the graph theory.
It can be understood that, for the constructed artificial intelligence application, after the server obtains the character string and decodes and converts the character string back to the dictionary, the server reconstructs mathematical language representation of each mathematical primitive on the graph theory based on the dictionary, so as to reconstruct the mathematical operation executed by the artificial intelligence application and the input and output dimensions of the executed mathematical operation rapidly.
The artificial intelligence application obtained by building a plurality of chunks is composed of a plurality of mathematical primitives, the newly added chunk forms a section of executable program, and similarly, the newly added chunk is also built by at least more than one chunk, therefore, the reconstruction of the mathematical primitives of the newly added chunk is also based on the reconstruction of mathematical language representation of each mathematical primitive on the graph theory by a dictionary.
After the mathematical primitive corresponding to the newly added chunk is obtained through reconstruction, updating on a graphical interface and a server side is needed. On one hand, a newly added chunk exists in a plurality of chunks which can be provided for a user to select, so that the user can add the newly added chunk to the building area, on the other hand, the server can configure the newly added chunk in the chunk initialization for the graphical interface, and identify data related to the newly added chunk in the received dictionary, so that mathematical primitives corresponding to the newly added chunk can be operated at the server side.
The newly added mathematical primitive updating includes updating the image information related to the chunk, and in addition, unique identification is carried out on the newly added mathematical primitive and mathematical primitive identification and core parameters are stored in all the mathematical primitives forming the newly added mathematical primitive, so that the updating of the newly added mathematical primitive also includes updating the mathematical primitive definition information such as the mathematical primitive identification and the core parameters.
The definition information of the mathematical primitive is updated at the user terminal and the server, and the mathematical operation executed by the newly added mathematical primitive and other information related to the mathematical operation can be obtained through the updated definition information of the mathematical primitive. The updated mathematical primitive definition information describes both the new mathematical primitive and the performance of the mathematical operations in the new mathematical primitive.
Through the exemplary embodiment, a function of newly adding the chunks is provided for the artificial intelligence application building, so that a user can build the chunks according to the application building requirement and the configuration chunks personalized to the habit, besides the original configured chunks, the user-defined configured chunks also exist on the graphical interface, and the convenience and the freedom of the artificial intelligence application building are enhanced.
In this exemplary embodiment, chunks that are frequently combined together for artificial intelligence application building are packaged together to form a new chunk. In other words, original more detailed mathematical operations are combined together to form a new mathematical operation, so that a new operation is built for the artificial intelligence application performed by the user, and the execution performance of the artificial intelligence application building is enhanced.
The new chunk updating process executed on the graphical interface comprises the following steps: and adding the newly added chunk to the graphical interface for the newly added mathematical primitive according to the configured chunk name so as to update the added newly added chunk in a plurality of chunks initially configured in the graphical interface.
For the newly added mathematical primitive, the graphical interface necessarily has a corresponding newly added chunk, that is, the newly added mathematical primitive is necessarily represented in the graphical interface in the form of the newly added chunk. In the graphical interface, the chunks representing the mathematical primitives, also called tiles, are the targets manipulated by the user for building the artificial intelligence application.
It should be appreciated that for user identification purposes, the newly added chunk will be configured for the newly added mathematical primitive on the graphical interface, and the newly added chunk will be marked with the corresponding chunk name to distinguish it from other chunks, thereby facilitating user selection.
Specifically, for adding a newly added chunk to the graphical interface according to the configured chunk name for the newly added mathematical primitive, so as to update the added newly added chunk in a plurality of chunks initially configured in the graphical interface, the corresponding execution process of this step includes:
generating a newly added mathematical primitive identifier for the newly added mathematical primitive;
generating mathematical primitive definition information for the mathematical primitive identifier and the core parameter packaged by the newly added mathematical primitive under the newly added mathematical primitive identifier;
and updating the mathematical primitive definition information to the server, wherein the updating of the mathematical primitive definition information at the server enables the newly added mathematical primitive to newly add a corresponding chunk in the initialization of the graphical interface at the server.
The newly added mathematical primitive identifiers, similar to the mathematical primitive identifiers mentioned above, are used to uniquely identify the corresponding mathematical primitive and chunk.
In the configuration of the core parameters for adding new mathematical primitives through adding new blocks, the configured core parameters correspond to the generated new mathematical primitive identifications.
The newly added chunk is formed by at least one chunk, correspondingly, a newly added mathematical primitive corresponding to the newly added chunk is formed by packaging and packaging at least one mathematical primitive, and for this purpose, at least one group of mathematical primitive identifications and core parameters are corresponding to the newly added mathematical identifier generated for the newly added mathematical primitive.
And describing the newly added mathematical primitive together with other sets of mathematical primitive identifications and core parameters, namely generating mathematical primitive definition information for the newly added mathematical primitive.
It should be understood that for the chunks existing in the graphical interface and available for the user to select, there are corresponding mathematical primitive definition information, so as to provide the corresponding algorithm implementation for the artificial intelligence application built for the user based on the mathematical primitive definition information.
The existence of the mathematical primitive definition information at the server side means that the corresponding newly-added mathematical primitives and chunks are deployed in the subsequent artificial intelligence application construction, the subsequent artificial intelligence application construction for the user through the graphical interface can select the newly-added chunks, and then the newly-added mathematical primitives are configured in the constructed artificial intelligence application.
In addition, in the implementation of mathematical primitive newly-adding on the user side, the process of generating the dictionary according to the chunk acted by the newly-added chunk selection instruction comprises the following steps:
respectively acquiring corresponding mathematical primitive identifications and core parameters according to all the chunks acted by the newly-added chunk selection instruction;
and for each chunk, constructing dictionary data by taking the corresponding mathematical primitive identification as an index item and taking the core parameter as an index value so as to generate a dictionary corresponding to the newly added chunk.
As mentioned above, the new chunk is implemented corresponding to at least one chunk selected by the user, and at least one mathematical basis group corresponding to the chunk is combined together to form a new chunk. Once the user selects the newly-added chunk, the corresponding at least one chunk is selected, so that the convenience and the efficiency of building the artificial intelligence application are enhanced, and the artificial intelligence application can be built quickly.
In view of at least one chunk selected to be newly added to the building area, the chunks are connected with each other to form a network topology capable of completing at least one mathematical operation, and at this time, since the chunks are linked with each other, input dimensions and output dimensions in core parameters corresponding to the chunks are already configured according to a currently constructed link relationship, and corresponding mathematical primitive identifications and core parameters can be obtained for each chunk so as to be used for generating dictionary data for each chunk, and the dictionary data indicates a link relationship between the chunk and other chunks linked thereto.
And constructing dictionary data for each chunk by using a data structure of the key value pair, wherein the data structure of the key value pair takes mathematical primitives as index items and core parameters as index values.
And constructing dictionary data for all the chunks which form the newly added chunk respectively, and generating a dictionary corresponding to the newly added chunk from all the dictionary data.
Just as chunk adding is carried out on a graphical interface of a user terminal, a server controlled by the user terminal inevitably receives mathematical primitive definition information sent by the user terminal, and at the moment, the mathematical primitive definition information is stored for the user so as to be called by the user in subsequent artificial intelligence application building.
Certainly, the chunk newly added by the user can also be shared with other users to be used in the artificial intelligence application building of other users, and is not limited herein.
The artificial intelligence application requirement of the user is the action recognition of the short video, and the method is combined to realize the exposition
The above-mentioned processes are described.
The user needs to build the artificial intelligence application to realize the video action recognition. Specifically, for a short video describing an action, the action can be recognized through an artificial intelligence application set up by a user, and an action classification result is output.
At a user terminal, a user can drag a required chunk on an operation component selection column of a graphical interface so as to drag the required chunk to the operation interface, and a link relation is established between the chunks through a connecting line until the construction of an artificial intelligence algorithm configuration is completed.
The neural network operation adopted in the constructed artificial intelligence algorithm configuration is mathematical operation adopted by the user to realize artificial intelligence application.
FIG. 9 is a graphical interface diagram shown in accordance with an exemplary embodiment. In an exemplary embodiment, as shown in FIG. 9, the operations component selection column 810 sets a number of operations components, i.e., chunks associated with data, models; the operation interface 830 is a stop area where the user drags the desired chunk, and data and model building is completed by dragging an icon, i.e., a component.
In the data and model building that is performed, fine tuning of the data and model will also be performed through the right toolbar 850.
The construction of artificial intelligence application is completed through the process, and iterative training of parameters is performed. Information related to the iterative training process, such as the number of iterations and the related iterative training results, is displayed through the operation result display area 870, and here, the training accuracy of the data and the model is also displayed, so that the user can be guaranteed to grasp the real performance of the artificial intelligence.
The method comprises the steps that an artificial intelligence application capable of realizing video action videos is built by a user, different artificial intelligence algorithm configurations can be tried through continuous chunk dragging, and an available artificial intelligence algorithm configuration can be found.
For example, a block corresponding to a convolutional neural network operation may be dragged to an operation interface from an operation component selection column 810 to construct a first layer of convolutional neural network, fig. 10 is a schematic diagram of the block corresponding to the convolutional neural network operation on the operation interface according to an exemplary embodiment, block I represents input, block Dense represents convolutional neural network operation in a deep learning neural network, block O represents output, on the basis, two blocks corresponding to the convolutional neural network operation are dragged to the operation interface to construct a three-layer convolutional neural network, so as to implement convolution feature extraction in the constructed artificial intelligence application, and fig. 11 is a schematic diagram of block distribution and link of the three-layer convolutional neural network on the operation interface according to the embodiment corresponding to fig. 10.
Through the interaction between the user terminal and the server, the user can continuously try to build the artificial intelligence application, and the effect is continuously verified, so that the operation is simple and the cost is low.
The realization of the invention helps the user to quickly realize the artificial intelligence application required by the user, and the artificial intelligence application can be called at will when required.
For a person with zero knowledge base related to artificial intelligence, the process of realizing the autonomous development of artificial intelligence application is roughly as follows: firstly, completely learning code knowledge, having a certain code development capability, then completely learning artificial intelligence professional knowledge, and finally, selecting a proper programming language as a developer to carry out algorithm development in a compiler according to the current specific requirements, wherein the time period consumed in the process is at least more than 7 years.
It should be understood that the code knowledge, deep learning algorithm and mathematical expression related to machine learning, and overall knowledge of artificial intelligence are difficult to popularize in the public, and the learning threshold of the code is difficult to exceed by the public.
In order to overcome the difficulties, the invention firstly provides a user-friendly interface for the public, packages a plurality of technical details in the form of chunks and mathematical primitives and realizes a standardized artificial intelligence model building process, thereby really helping the public to use the artificial intelligence technology.
Artificial intelligence algorithms are a complex natural expression that is difficult to implement by means of graphical user programming, unlike simple algorithmic logic similar to that involved in child programming.
In the invention, the code encapsulation is realized through the mathematical primitives so as to realize the required mathematical operation, and the graphic blocks are defined on the graphic interface for representation.
Fig. 12 is a schematic diagram illustrating interactions between a front end and a back end in accordance with the present invention, according to an exemplary embodiment. In this exemplary embodiment, the user side implements a website graphical interface, that is, for the user, the required artificial intelligence application is implemented through a web application, and the web application can be run through the browser only by acquiring the access address corresponding to the background.
Of course, the present invention is not limited to this, and may be realized in many ways such as by issuing a terminal program.
As shown in fig. 12, through the manipulation of the user on the graphical interface of the website, the server in the background converts the chunks, i.e., the graphs, configured by the user into codes, so as to obtain the constructed artificial intelligence application, and returns the result to the user by means of the cloud computing capability provided by the server, or the artificial intelligence algorithm configuration graph. On a graphical interface of a website, a user constructs an AI (Artificial Intelligence) algorithm by using a mathematical language representation of graph theory, namely the mathematical primitives referred to above, and related algorithm information is stored in a dictionary realized by a JavaScript language. On one hand, a user selects to train the algorithm on a website graphical interface, then the dictionary can be submitted to a server in a background in the form of JSON character strings, the server decodes the JSON character strings from an output end to an input end into an executable AI code language after receiving the JSON character strings, and the server completes training according to the AI code language and returns the result to the website graphical interface for displaying.
It should be noted that the Mathematical primitive corresponds to a Mathematical operation, and the Mathematical operation may include Basic Mathematical Operations such as addition, subtraction, multiplication, division, etc. (BMO), and various complex Mathematical Operations such as neural Network Operations, etc., and the Mathematical language expression of the Mathematical primitive in graph theory is a DAG structure for expressing the Mathematical operation, also called mng (Mathematical Network graph). A dag (directed Acyclic graph) structure for describing application logic representing expressions in the form of nodes.
Fig. 13 shows a schematic diagram of the implementation of the overall scheme of the present invention in an exemplary embodiment. In one exemplary embodiment, as shown in FIG. 13, for an artificial intelligence application autonomously developed by a user, it includes four major phases, namely: the build phase, the send graph phase, the translate dictionary to MNG phase and the return result phase.
The construction period is a period of constructing the required artificial intelligence application by the user, and at this period, the user can carry out the artificial intelligence algorithm and the construction of a new MNG block by the chunk, and the corresponding mathematical primitive index and the core parameter are marked and stored in the dictionary no matter the construction of the artificial intelligence algorithm or the construction of the new MNG block.
The send graph phase is performed by the build period. The map sending stage is a stage of sending a dictionary for the constructed artificial intelligence algorithm or the new MNG block, and is called a map sending stage, which means that the artificial intelligence algorithm or the MNG block corresponding to the dictionary corresponds to a network topology map.
The type conversion of the character strings received by the server is realized in the graph sending stage, so that the character strings are eliminated and are decoded and converted back to the dictionary.
The send graph phase enters the translate dictionary to MNG phase, at which the server will perform recognition of MNG, i.e. mathematical primitives, to generate code through DAG structure, obtain executable AI code language, train and return results.
The trained AI code language can be used to implement decisions on the data. The decision phase is implemented on the data, still running and returning the results. In the process, after the server completes the method construction of the artificial intelligence application, the AI code language is obtained, at this time, the model is identified and whether the model is trained or not is checked, if the model is trained and contains data to be run, the AI code language is run by using the data, and the result is returned.
Optionally, before the decision is made on the data, the user is confirmed to want by showing the artificial intelligence algorithm configuration diagram to the user, on the basis of which the previously trained AI code language can be run.
The artificial intelligence algorithm pattern is obtained by block construction when a user builds an artificial intelligence application. The process of obtaining AI codes is the process of extracting important information from the artificial intelligence algorithm configuration diagram and analyzing the information into codes, and packaging is carried out according to the important information, namely packaging is carried out to each data element, and net-to-gate, namely, a network structure is changed into a single mathematical operation, is realized. By now it should be appreciated that any mathematical operation or mathematical equation is a complex of BMOs that are linked together to form a DAG structure, which is expressed as a MNG. Each MNG has respective input and output, and the input and output between different MNGs can be mutually linked to form a complete logic network, so that the input and output can be used as an equivalent transition form between the chunks and the codes, and the information corresponding to the chunks can be accurately transmitted to the codes by means of the equivalent transition form, and the conversion between a graphical interface and a text programming language is realized.
It should be understood that each of the mathematical primitives is implemented at the beginning, and at the same time they are the basis for the user's manipulation, and the network structure built by the user through the chunks will be converted into a dictionary as before. The dictionary is then converted to a DAG format and parsed. The network structure then becomes a mathematical primitive and is then added to the server as a new mathematical primitive. The newly formed mathematical primitives can be used as well as other primitives to construct other networks. As this loop is repeated, more complex algorithms can be developed without increasing the difficulty of graphic base programming language development.
The following is an embodiment of the apparatus of the present invention, which is used for implementing an embodiment of an operation implementation apparatus in the artificial intelligence application building of the present invention. For details that are not disclosed in the embodiment of the apparatus of the present invention, please refer to the embodiment of the operation implementation method in the artificial intelligence application building of the present invention.
FIG. 14 is a block diagram illustrating an apparatus for implementing operations in the construction of an artificial intelligence application, according to an example embodiment. In an exemplary embodiment, as shown in fig. 14, the apparatus for implementing operation in the building of artificial intelligence application includes: a receiving module 1010, a dictionary obtaining module 1030, a text generating module 1050, and an application running module 1070.
A receiving module 1010, configured to receive an artificial intelligence algorithm configuration selection performed by a user on the constructed artificial intelligence application, where the artificial intelligence algorithm configuration is a network topology describing mathematical operations performed in the artificial intelligence application;
a dictionary obtaining module 1030, configured to select and obtain a dictionary describing the executed mathematical operation according to the artificial intelligence algorithm configuration;
the text generation module 1050 is configured to reconstruct a mathematical operation on the dictionary through a mathematical language representation of graph theory to obtain an executable text for executing the mathematical operation;
and the application running module 1070 is configured to enable the user to run the built artificial intelligence application according to the selected artificial intelligence algorithm configuration through the execution of the executable text.
Optionally, the present invention further provides an electronic device, which may be used in the implementation environment shown in fig. 1 to execute all or part of the steps of the method shown in any one of fig. 3, fig. 4, fig. 5, fig. 6, fig. 7, fig. 8, and fig. 9. The device comprises:
a processor;
a memory for storing processor-executable instructions;
wherein the processor is configured to execute the method for realizing the foregoing.
The specific manner in which the processor of the apparatus in this embodiment performs operations has been described in detail in relation to the foregoing embodiments and will not be elaborated upon here.
In an exemplary embodiment, a storage medium is also provided that is a computer-readable storage medium, such as may be transitory and non-transitory computer-readable storage media, including instructions. The storage medium includes, for example, the memory 204 of instructions executable by the processor 218 of the device 200 to perform the methods described above.
It will be understood that the invention is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the invention is limited only by the appended claims.
Claims (10)
- An operation implementation method in artificial intelligence application building is characterized by comprising the following steps:receiving artificial intelligence algorithm configuration selection of a user on the constructed artificial intelligence application, wherein the artificial intelligence algorithm configuration is a network topology for describing mathematical operations executed in the artificial intelligence application;selecting and obtaining a dictionary describing the executed mathematical operation according to the artificial intelligence algorithm configuration;reconstructing mathematical operation on the dictionary through mathematical language representation of graph theory to obtain executable text for executing the mathematical operation;and through the execution of the executable text, the user is enabled to run the built artificial intelligence application according to the selected artificial intelligence algorithm configuration.
- The method of claim 1, wherein receiving an artificial intelligence algorithm configuration selection by a user of the constructed artificial intelligence application comprises:receiving and obtaining an artificial intelligence algorithm configuration selection of the constructed artificial intelligence application by the user according to the selection of the constructed artificial intelligence application by the user, wherein the artificial intelligence algorithm configuration selection is mapped with a dictionary stored for the artificial intelligence application.
- The method of claim 1, wherein receiving an artificial intelligence algorithm configuration selection by a user of the constructed artificial intelligence application comprises:and according to the artificial intelligence application construction selected by the user, receiving a character string corresponding to the dictionary, wherein the dictionary corresponding to the character string is used for recording the artificial intelligence algorithm configuration formed by the artificial intelligence application constructed by the user through mathematical primitive configuration at present.
- The method of claim 3, wherein said selecting a dictionary describing mathematical operations performed according to said artificial intelligence algorithm configuration comprises:and decoding and converting the character string into a dictionary to obtain the dictionary formed by taking the digital element identification and the core parameter as dictionary data, wherein the digital element identification and the core parameter in the dictionary data correspond to the mathematical element for building the artificial intelligence application configuration.
- The method of claim 1, wherein reconstructing a mathematical operation on the dictionary through the graph-theoretic mathematical language representation results in an executable text that performs the mathematical operation, comprising:acquiring data missing codes of corresponding mathematical operations according to the mathematical primitive identifications in the dictionary;filling core parameters corresponding to the mathematical primitive identification into the obtained data missing codes to obtain code information for executing corresponding mathematical operations;and according to the input dimension and the output dimension indicated in the core parameter corresponding to the mathematical primitive identification, sequentially reconstructing the executable text of the artificial intelligence application through the mathematical language representation of graph theory from output to input of the code information.
- The method of claim 1, wherein the causing the user to run the built artificial intelligence application according to the selected artificial intelligence algorithm configuration through the execution of the executable text comprises:acquiring sample data suitable for the artificial intelligence application in a training mode, wherein the training mode is an executed artificial intelligence application running mode;and performing parameter training through the sample data in the operation of the artificial intelligence application through the artificial intelligence application set up by the execution operation of the executable text.
- The method of claim 6, wherein the causing the user to run the built artificial intelligence application according to the selected artificial intelligence algorithm configuration through the execution of the executable text comprises:acquiring data selected and input by a user, wherein the data comprises target data processed by an artificial intelligence application set up by the user;and running the trained artificial intelligence application in a decision mode to finish the processing of selecting input data by a user by executing the executable file finishing parameter training.
- The method of claim 1, wherein the mathematical operations that can be performed correspond to a mathematical primitive and the artificial intelligence algorithm configuration is a network topology formed by the interlinked mathematical primitives, the method further comprising:receiving the update of a user to a newly added mathematical primitive to obtain mathematical primitive definition information, wherein the mathematical primitive definition information comprises a newly added mathematical primitive identifier, at least one mathematical primitive identifier and a core parameter, and the at least one mathematical primitive identifier and the core parameter correspond to the newly added mathematical primitive encapsulated under the newly added mathematical primitive identifier;and storing the mathematical primitive definition information to initialize and configure the graphical interface with the newly added chunk.
- An operation implementation device in artificial intelligence application building, characterized in that the device comprises:the receiving module is used for receiving artificial intelligence algorithm configuration selection of a user on the constructed artificial intelligence application, and the artificial intelligence algorithm configuration is a network topology for describing mathematical operations executed in the artificial intelligence application;the dictionary acquisition module is used for selecting and acquiring a dictionary describing the executed mathematical operation according to the artificial intelligence algorithm configuration;the text generation module is used for reconstructing mathematical operation on the dictionary through mathematical language representation of graph theory to obtain an executable text for executing the mathematical operation;and the application running module is used for enabling the user to run the built artificial intelligence application according to the selected artificial intelligence algorithm configuration through the execution of the executable text.
- A machine device, comprising:a processor; anda memory having computer readable instructions stored thereon which, when executed by the processor, implement the method of any of claims 1 to 8.
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2018/125254 WO2020133323A1 (en) | 2018-12-29 | 2018-12-29 | Method and device for running implementation in construction of artificial intelligence application, and machine equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111868683A true CN111868683A (en) | 2020-10-30 |
| CN111868683B CN111868683B (en) | 2023-11-17 |
Family
ID=71126785
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201880002680.7A Active CN111868683B (en) | 2018-12-29 | 2018-12-29 | Operation realization method, device and machine equipment in artificial intelligent application construction |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111868683B (en) |
| WO (1) | WO2020133323A1 (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103472756A (en) * | 2013-09-27 | 2013-12-25 | 腾讯科技（深圳）有限公司 | Artificial intelligence achieving method, server and equipment |
| CN107391111A (en) * | 2017-06-22 | 2017-11-24 | 刘武丰 | Artificial intelligence co-development framework and implementation method |
| US20170357893A1 (en) * | 2016-06-10 | 2017-12-14 | Apple Inc. | Artificial intelligence controller that procedurally tailors itself to an application |
| CN107908743A (en) * | 2017-11-16 | 2018-04-13 | 百度在线网络技术（北京）有限公司 | Artificial intelligence application construction method and device |
-
2018
- 2018-12-29 WO PCT/CN2018/125254 patent/WO2020133323A1/en active Application Filing
- 2018-12-29 CN CN201880002680.7A patent/CN111868683B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103472756A (en) * | 2013-09-27 | 2013-12-25 | 腾讯科技（深圳）有限公司 | Artificial intelligence achieving method, server and equipment |
| US20170357893A1 (en) * | 2016-06-10 | 2017-12-14 | Apple Inc. | Artificial intelligence controller that procedurally tailors itself to an application |
| CN107391111A (en) * | 2017-06-22 | 2017-11-24 | 刘武丰 | Artificial intelligence co-development framework and implementation method |
| CN107908743A (en) * | 2017-11-16 | 2018-04-13 | 百度在线网络技术（北京）有限公司 | Artificial intelligence application construction method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020133323A1 (en) | 2020-07-02 |
| CN111868683B (en) | 2023-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11922322B2 (en) | Exponential modeling with deep learning features | |
| JP7208952B2 (en) | Method and apparatus for generating interaction models | |
| US20200242771A1 (en) | Semantic image synthesis for generating substantially photorealistic images using neural networks | |
| Paternò | End user development: Survey of an emerging field for empowering people | |
| US11087518B2 (en) | Second-person avatars | |
| CN105706033B (en) | System and method for cube graphic user interface | |
| US10515119B2 (en) | Sequential recommender system for virtualized network services | |
| KR102131159B1 (en) | Methods and devices for controlling smart devices and computer storage media | |
| US11604641B2 (en) | Methods and systems for resolving user interface features, and related applications | |
| JP2020520504A (en) | Deep Adaptive Semantic Logic Network | |
| Lande et al. | GPT Semantic Networking: A Dream of the Semantic Web–The Time is Now | |
| US10762902B2 (en) | Method and apparatus for synthesizing adaptive data visualizations | |
| US20210150391A1 (en) | Explainable Machine Learning Predictions | |
| KR102213291B1 (en) | System for making website | |
| Gore et al. | Django: Web development simple & fast | |
| CN111819536A (en) | Method, device and machine equipment for realizing construction and operation of artificial intelligence application | |
| CN113077237A (en) | Course arrangement method and system for self-adaptive hybrid algorithm | |
| CN111868683B (en) | Operation realization method, device and machine equipment in artificial intelligent application construction | |
| US20230186117A1 (en) | Automated cloud data and technology solution delivery using dynamic minibot squad engine machine learning and artificial intelligence modeling | |
| CN111712791B (en) | Method, device and machine equipment for newly adding blocks in artificial intelligence application building | |
| US20230297625A1 (en) | Utilizing a graph neural network to generate visualization and attribute recommendations | |
| KR102549937B1 (en) | Apparatus and method for providing model for analysis of user's interior style based on text data of social network service | |
| Boyer-Kassem | Layers of models in computer simulations | |
| KR20220044264A (en) | Systems and methods for deploying low-application-impact user interfaces | |
| CN111967570B (en) | Implementation method, device and machine equipment of visual neural network system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |