WO2020236021A1

WO2020236021A1 - Universal and personalized distributed decision support system

Info

Publication number: WO2020236021A1
Application number: PCT/RS2019/000016
Authority: WO
Inventors: Gordana VELIKIC; Milan VIDAKOVIC; Ivan Kastelan; Nikola Teslic
Original assignee: Rt-Rk Istrazivacko-Razvojni Institut Rt-Rk Doo Za Sisteme Zasnovane Na Racunarima Novi Sad
Priority date: 2019-05-17
Filing date: 2019-05-17
Publication date: 2020-11-26

Abstract

A universal and personalized, distributed decision support system (100) comprising of: a script machine (101) that processes data of users' objects of the application servers (105) users by calling the library (108) of functions, a predictive model (102) which determines the shape and function coefficients adapted to the users of the application servers (105) and their objects, and personalized models (103) that communicate with the aforementioned system (100) through the middle layer software support (104) have, for the novelty, a feature of universality which is realized by the implementation of plug-in (109) for the various deployment of the system (100) through which the system 100 communicates with function libraries (108). Plug-in (109) is used to customize data and functions when calling the function library (108). Second feature of the system (100) is that it is personalized, which is reflected in the fact that the personalized models (103) contained in the system 100 are assigned to users and their objects, wherein personalized models (103) are formed by integrating said predictive model (102) with measurements acquired from objects of end users of application servers (105) and coefficients of the predictive models (102). The coefficients are adapted to the specified objects of the current users of the application servers (105). The user can be a human, an animal, a plant, a system, a device, an artificial intelligence, etc. while object can be a field, a vehicle, a human, etc., while variables are for example, humidity, pressure, temperature, result of the survey, etc. Different applications of the system (100) can be health, meteorology, agriculture, industry, etc. Application servers (105) reflect different deployments of the system (100).

Description

UNIVERSAL AND PERSONALIZED DISTRIBUTED DECISION SUPPORT

SYSTEM

Technical Field

The invention is related to the field of interactive computer information systems which process large amount of data and with the aid of plug-ins and methods generate decisions that are in the user’s best interest. Additional contributions of the invention are the user-tailored decisions, i.e. personalized decisions, and simple application of the entire decision support system (DSS) to different fields, such as but not limited to agriculture, health, different industries, education, and consumer applications.

According to international patent classification, the application belongs to the following IPC groups: G06N20/00 H H04L69/321.

Background Art

The invention contributes to the new decision support system because it solves the problem of simple application of the decision support platform in various fields of application in industry and economy through the software plug-in. The invention solves the problem of additional software configuration for different applications of the DSS. It also provides a novel concept of personalization of the platform itself, which increases efficiency and speed.

To optimize the work process and to solve the problem of speed, time, knowledge, and errors mainly during writing of scripts by programmers, the invention introduces a novel solution where is developed a plug-inable DSS, which - for each application (e.g., agriculture, healthcare ...) where a specific library and function library call is required (for example, the Tensor Flow platform) - activates a separate plug-in, thus achieving the speed and efficiency of the DSS, that is - universality rather than a creation of a new script for every change in the system.

Some of the protected solutions and research papers that represent a prior art of the proposed invention are listed below.

Patent application US20170213132 A1 published on January 27^th, 2016 entitled "Multiple user interfaces of an artificial intelligence system to accommodate different types of users solving different types of problems with artificial intelligence" provides a solution for applying a platform configured to interconnect with a large number of users is adapted to different types of users in order to solve various types of problems using artificial intelligence. The artificial intelligence mechanism can include artificial intelligence modules as well as architecture modules, instruction modules and training modules. The solution does not have a universal and personalized decision support platform (DSS) in a way that suggests the proposed invention. Patent application US20170006141 A1 published February 7^th, 2015 under the title "Cognitive Intelligence Platform for Distributed M2M / IoT Systems" provides a system and method for managing data and knowledge during communication between intelligent devices. The cognitive intelligent platform is designed for the IoT-Intemet of Things network and provides automated support when deciding in real time, in a dynamic work environment. The data cognitive intelligence platform has three logical layers of data processing with increased complexity, each of which has an agent using statistical, and engine learning techniques and algorithms for solving various problems and improving knowledge. The above application is a general learning model, a more similar solution to the server of recommendations, while the proposed invention introduces the personalization and universality of the decision support system (DSS).

Patent US6859798 Bl, published June 20^th, 2001, entitled "Intelligence Server System", describes an intelligent server which generates reports. The reporting system includes a report initiation module such as a user block or a calling application which generate report requests. An intelligent server includes: a server layer for receiving the generated report and for coordinating the received request stream; an application logic layer that includes business intelligence rules used to generate reports; and an analytical layer in communication with the user block and the server layer for receiving requests to be processed by the server layer. The given invention belongs to areas for processing and analyzing data, more precisely for the scalable use of the Business Intelligence Platform.

Patent US7181493 B2 published December 23^rd, 2003 entitled "Platform Independent Model- Based Framework for Information Exchange in the Justice System" provides an independent platform for the information exchange of a large number of different entities in the legal system with heterogeneous components implemented in a hub and spoke arrangement. Each spoke has a software agent that communicates with the system component of the entity's legal system, translates from a common communication format to the communication format of the system component, and manages time of the information flow to and from the system component. An independent platform model includes a large number of linked models that define broker information from the used business model to an independent platform model to implement the given solution. The mentioned patent does not mention the prediction model and the use of neural networks, nor the possibility of applying it to several different systems at the same time, i.e. universality.

Patent US6151582 published November 21^s1, 2000 under the title "Decision Support System for the Management of an Agile Supply Chain" provides architecture that includes a client and server side. The server side includes a DSS database that is associated with a modeling device that analyzes information for the decision support planning. The server side also includes a server manager that coordinates the requests. The client side includes frames for displaying some available points in the user system. The frame manager coordinates the requests to access the necessary data and modules. The invention belongs to the support system which organizes decisions in the supply chain and thus differs from the proposed idea that supports different industrial areas at the same time, that is, the universality. Patent application US20140057255 A1 published September 25^th, 2011 under the title "Systems and Methods for Collecting and Transmitting Assay Results" provides a solution for the acquisition, preparation, and organization of health data of individual patients. Collection of samples is done using a sampling device. The sample processing device performs sample preparation and / or chemical reaction related to the samples. Data related to samples can be forwarded from the data acquisition unit to the laboratory. The laboratory can be a certified laboratory that generates a report that can be forwarded to a health center. A health center may rely on the diagnosis, treatment and / or prevention of a disease report of an individual. The present invention is applied in the field of medicine and there is no possibility of simultaneous application in other areas.

The present state of the art also includes the following patent applications and patents: patent US6954757 titled "Decision-management system which is cross-function, cross-industry and cross-platform" published on July 29^th, 1999, then patent application US20020091687 titled "Decision support system" published on July l l^lh, 2002, then US20040215546 patent application titled "Systems and methods for investment decision support" published July 11^th , 2002, then US6699193 patent titled "Decision Support Systems and Methods for Assessing Vascular Health" published May 23^rd , 2002 as well as patent application US20050209732 titled "Decision support system for supply chain management" published September 22^nd 2005, US20110301977 patent application titled "Systems and methods for value-based decision support" published December 8^th, 2011, patent application titled "System and method for providing educational related social / geo / promo link promotional data sets for the end user display of interactive ad links, promotions and sales of products, goods, and / or services integrated with 3d spatial geomapping, company and local information for selected worldwide locations and social networking" published October 9^th, 2011 and patents US6859798 titled “Intelligence Server System”, published February 22^nd, 2005 and US8260736 titled “Intelligent System Manager System and Method”, published September 4^th, 2012. All of these solutions essentially differ from the invention in part that their solutions DSS is not universal or pluggable (a universality ensurement accessory) for various applications. Also, personalization related to users and their objects is not listed.

Disclosure of the Invention

The Decision Support System (DSS) supports the decision-making process along with the activities aimed at resolving decision-making issues. The DSS activities include system modeling, and the DSS itself facilitates the decision-making process to create the most effective output, while assuring that the important details of the user side, the application field, and the type of input data are not neglected.

To obtain the high precision quality of models and decision making processes in such systems, often are used linear and nonlinear regressive models, which include, inter alia, polynomial function based models, and Artificial Neural Networks (ANN).

Neural networks represent a mathematical model based on biological neural networks. The ANN is based on artificial neurons that are interconnected, with intrinsic adaptive character, so it can easily change the structure based on the information transmitted through the network during the training phase.

The process that the invention proposes is based on monitoring the state of the various users of the application servers in real time. The user can be a human, another system, entities from a plant or animal world, a device, a form of artificial intelligence such as a robot, and the system itself processes data that it receives from objects in the form of variables. Objects can be cars, lots, people, animals, plants, other systems, artificial intelligence entity etc., while changing variables can be humidity, temperature, pressure, velocity, etc. Modeling is performed for each object of application servers with ultimate goal to predict the future variables of these objects.

The invention proposes that data received from end users (or their objects) from different application sources (application servers) further via middleware is sent to the DSS platform, which contains a part that executes different scripts and processes the acquired data. To avoid writing scripts for each application area, the DSS platform uses plug-ins to invoke platfonns with function libraries, such as for example, Tensor Flow, which in case of invention provides adequate processing function for current data. This adequate processing function again in the case of the invention can be a linear or nonlinear regression, that is, a polynomial of functions, and generally can be, for example, a neural network or some other function. The DSS predictive model forms the function it receives from the library of functions (for example, it can be a polynomial function to which then DSS elements calculates the coefficients and the order of the polynomial). The initial prediction model is a general model and when real measurements are made by the end users with the coefficients of the predictive model, then it becomes a personalized model. Personalization is reflected in the fact that personalized models are assigned to application servers and that they were created by integrating the predictive model with the measurements and coefficients of the predictive model of the mentioned function or the specified polynomial. The application server user can be a human or system, a device, a form of artificial intelligence such as a robot, an animal like, for example, a pet, that is, one that can react to the exits from the described system, while the object can be a man and, for example, fields, cars, pets, etc., and variable conditions can be humidity, temperature, speed, pressure, etc. Application servers reflect the various applications of the said DSS.

The coefficients of the predictive model function in the form of operational data are stored by DSS in its internal memory. This add-on which enables access to the library of functions (such as for example Tensor Flow platform) for any application server, makes the universality of the system of invention. The above mentioned personalization feature is also an innovative feature of the DSS system that the invention proposes. Also, the invention proposes the distribution of system components because the function library is separate and communicates with the DSS system components via the add-on. Brief Description of the Drawings

Figure 1 - shows the basic concept of the invention

Figure 2 - shows the general implementation of the components of the invention

Best Mode for Carrying Out of the Invention

The invention introduces a new concept of the DSS. The decision support system itself is a set of computer programs and data used to assist in analysis and decision-making in various applications of the system. The computer program analyzes data and returns results to end users so that they can make easier decisions. Often these decisions are made by the trained network of the support decision system itself such as neural network-NN network, which is implemented in a model-structured DSS through a predictive model. The decision support system can include artificial intelligence. Big data that such system supports, are processed with a unique goal to solve and apply the solution of a specific challenge. It is always the ultimate goal to reach a decision that is the closest to current or future approximations. The decision support system of the invention is programmed to generate different types of output reports and decisions which are based on different types of users and different fields of application of the platform. Precisely this feature of universality (applicability for various fields of application-users of different application servers) is achieved with additional small programmability in the form of the so-called plug-in supplements which for different fields of application adjusts data with functions of predictive models used to process the data.

Application with such pluggable solution of the DSS can be deployed, for example, in agriculture, healthcare, automotive industry, etc.

The database contained in the aforementioned decision support platform, contains internal data from users and its objects with the measurements and personalized coefficients of the predictive model functions, stored on the application servers, and the operational data stored in the internal memory of the DSS. The decision support system contains various mathematical and analytical models that are used to analyze complex data. A modeled organized platform for decision support provides an output based on different inputs and different conditions. Each model determines a specific function where the selection of the model depends on the user's requirements and the puipose of the DSS.

Generally, the DSS can contain different models: a statistical model which may again carry statistical functions, a sensitively-analytical model that can contain analytic functions (often important for finance), then an optimization analytical model where the optimum value for a specific target is sought, then forecast models where future plans are made, etc.

In the case of the invention, a modular frame with a script engine is also provided as a basic model of script processing, then a predictive model that contains a general code for a function that is called externally from the function library by means of add-ons that depend on the field of application, and finally a personalized model that was created by the integration of the predictive model with actual measurements by the user and its objects and personalized coefficients of the given function (e.g. the polynomial function). In addition to universality and personalization, the DSS system is distributed in the sense that it has separated components, the decision support platform is separated from the function library and they communicate through the said add-on.

A special plug-in adapts the data received by the DSS platform with a function library (e.g. Tensor Flow). In addition to Tensor Flow, Java Coffee or some other function library can also be used. This DSS architecture through the models gives it a real-world abstraction, a representation of a system being explored, a simplification of reality. With them we can show as real a system as possible, that is, to better understand the system. Tensor Flow is an open platform, a program that is free to everyone, i.e. a software library for differential programming, for neural networks. This platform contains many algorithms and libraries, is of a flexible architecture, and is suitable for demanding analyzes through different domains of application. For the sake of the higher efficiency of implementation, the DSS calls library functions, such as for example, but not limited to Tensor Flow, via add-ons. This way of communication is a faster and more efficient solution compared to the previous writing and execution of scripts for each application of the DSS individually. The invention proposed approach increases the speed, simplicity, and reduction of errors.

The communication through the said appendix proposes a new structure of the DSS. Figure la shows the implementation of the decision support system 100-DSS system (platform) in accordance with the idea of the invention. The decision support system 100 is universal, personalized, and distributed. Universality is an innovative feature of the invention, because for various applications it provides a new way of communication with the library of functions 108 through plug-in 109. Personalization is reflected in the fact that personalized models 103 are related to the end user of the application servers 105 and the objects of those users.

In Figure la, a core of the DSS system is represented, which is a script engine 101 that executes the scripts, then the predictive model 102 which communicates with library 108 of functions through plug-in 109 and applies a dedicated function derived from the library functions 108 for the given application of the platform 100 and the type of data of the objects of users of the application servers 105 and the personalized model 103 which determines the shape of the function of the predictive model 102 and implements the actual parameters acquired via measurements from the final users of the application servers 105, and related to the objects of these users. The predictive model 102 can use a linear or nonlinear regressive model that contains neuron networks or a polynomial functions obtained by calling the library of 108 functions. Predictive model 102 determines the shape of the function in the way that in the case of the invention indicates the order of the polynomial function, and then the personalized model 103 inserts actual measurements related to the users of the application servers 105 and their objects from the field and also inserts the personalized coefficients of the predictive model which are linked to end users and their objects. Personalized models 103 are individually linked to 104 middleware and to objects of the end- users of application servers 105 which have 1, 2, ... n. Users have their own objects (e.g. plots, cars ...). DSS platform 100 connects, for example, with Tensor Flow or Java Coffee Platform 108 through plug-in 109, thus achieving the aforementioned universality and plug-ability of the invention. Adjustment of data via plug-in 109 as well as the call of a specific function from the library 108 functions depend on the application of the DSS platform 100 (application servers 105).

Neural networks, if applied in the predictive model 102, are trained to provide a precise prediction for the objects of the users of the application server 105. Part of the database 106 is located on the user’s side, and the DSS itself keeps own operational data. The communication between the end user and its objects and the DSS platform 100 also takes place through the input / output portion 107 on the user side of the application servers 105.

A translator, that is, the middleware 104 software support provides the transcription of generic data of objects of users of application servers 105. The personalization feature is described in the following example. If we observe a lot in agriculture as a system’s object, it is characterized by some state variables: temperature, humidity, etc., and as such it belongs to a user, whiles the coefficients and measurements from this lot are subject to personalization. Lots can be viewed individually and grouped, and a new user should take general information from the previous objects (lots), then fine-tune them because they personalize them exactly for that new lot whose data and decisions should be implemented. At the beginning, we have initial setup, while personalization for the chosen lot follows the initial setup. As time progresses with more inputs, the accuracy is rising. The initial setup takes place to ensure better accuracy before data intrinsic to the chosen lot are accumulated. Agriculture in this case is an application server, the human in this case is a user, and lots are objects, while the variable states indicate humidity, temperature, etc. with values measured in the field and envisaged model.

Figure lb represents further generalization of implementation of the DSS system (platform) 100 consisting of: an application server 105 which is a user side (with objects), then script engine 101 that is located in the DSS 100 itself, middleware 104, i.e. medium layer support software that communicates between the application server 105 and the DSS 100. The DSS 100 calls library functions, for example, Tensor Flow platform 108 and scripts 110 that are written in the Java Script programming language and execute by the engine script 101. To prevent execution of long scripts in DSS 100, i.e. wilting and execution of long scripts for script engine 101 for each different application of DSS 100 platform, the invention contributes to technical solution of the problem by activating only function libraries 108, (such as for example Tensor Flow, Java Coffee) and adapts them to a particular deployment of DSS 100. This exact adaptation is done by plug-in 109, in case of the invention and for different applications to prevent long codes. Scripts 110 represent scenarios data, macro commands, i.e. program to automate data.

The invention can use linear or nonlinear regression for data processing, which it calls from the library 108 of functions.

Script engine takes data from middleware 104, processes them (for example applies linear regression) and returns the results to middleware 104, which is called translator” to describe the role it has in DSS 100, and if needed returns the results to the users 105. The results are also stored on the DSS 100. Further iterations are executed while looking for the optimal function of the predictive model. When the iteration is completed, a function that has to be approximately similar to the real data is obtained, and based on this function, a prediction of future parameters can be made depending on the different source of application.

Plug-in 109 adjusts the communication between the DSS 100 and the library 108 of functions in a way to adjust the input / output parameters and it is a function of the various applications of the DSS platform (e.g. meteorology, agriculture, etc.).

Industrial Applicability

The invention for its multiple functionality can be deployed for different systems in agriculture, health, automotive industry, industry in general, i.e. wherever processing and analysis of large amounts of data is required, with the application of support and decision systems.

Claims

CLAIMS:

1. A universal and personalized, distributed decision support system 100 which contains: a script engine 101 which processes the objects data of the users’ of applications servers 105 by calling library 108 of functions, a predictive model 102, and a personalized model 103 which with the user of application servers 105 communicates via middleware software support 104, characterized by that the universal decision support system (100) utilizes the plug-in (109) for adjusting communication with the various application servers (105) of the system (100) using the library (108) of functions, and wherein the personalized models (103) are assigned to the users of the application servers (105) and to their objects.

2. A system according to the claim 1, characterized by the predictive model (102) is a liner progressive model.

3. A system according to the claim 1, characterized by the predictive model (102) is an artificial neural network.

4. A system according to the claim I, characterized by the predictive model (102) is a polynomial function.

5. A system according to the claim 1, characterized by as a predictive model (102), determines the shape of a function and personal coefficients of the said function.

6. A system according to the claims 1, 2, 3, and 4, characterized by the personalized coefficients of the predictive model (102) are adjusted to the users of the application servers (105) and their objects.

7. A system according to the claim 1 and 5, characterized by that the personalized models (103) are generated by the integration of the predictive models (102) and the personalized coefficients that are stored in the database (106).