Summary of the final work
When writing this essay master's work is not yet completed. Final Completion: June 2019. Full text of the work and materials on the topic can be obtained from the author or his manager after the specified date.
Content
- Introduction
- 1. Relevance of the topic
- 2. The purpose and objectives of the study, the planned results
- 3. Review of research and development
- 3.1 Overview of international sources
- 3.2 Overview of national sources
- 3.3 Overview of local sources
- 4. Agent-based modeling of the learning process of students at the level of competencies
- Findings
- List of sources
Introduction
Initially, the idea of an intellectual intermediary ("agent") arose in connection with the desire to simplify the end-user communication style. with computer programs. This idea proved to be very fruitful, and its implementation led to the creation of so-called "autonomous agents", which provided a new style of user interaction with the program. New was the fact that both the user and the computer agent-agent participated in the task launch, event management and problem solving. Research and experimental software development is pretty quickly showed that the multitude of tasks in which an agent can effectively assist a user is almost unlimited. These include searching for information on the Internet and its analysis, e-mail management, scheduling, e-commerce, assistance in any choice (preparation of recommendations for the selection of books, movies, music), etc. [1].
1. Relevance of the topic
In modern conditions, there is a growing need for analyzing and managing complex socio-economic and production systems, the state of which in most cases is unpredictable and cannot be predictable analytically from the start, because it is the result dynamic interaction of many heterogeneous active elements of the system and the environment.
Evaluation of the quality of education is an important task of school management. The complexity of its solution is due to the fact that the educational the processes are very slow, i.e. the training system as an object of management is inertial, therefore the effectiveness of innovative Changes can only be assessed after 4–6 years (student learning cycle). Using the simulation method, you can get a forecast in a short time the feasibility of innovations.
Master's thesis is devoted to the development of a simulation model that is focused on forecasting using a multi-agent approach. the quality of student learning in individual disciplines. For this, artificial agents of the simulation model must be delegated authority subjects of the educational process, i.e. functional responsibilities of students and teachers.
2. The purpose and objectives of the study, the planned results
Purpose of the study is the development of an agent-based model of student learning at the level of competence.
The work provides the solution of the following tasks:
- Formalization of forecasting the level of competence of students after completion of the course.
- Compilation of logical models describing the powers of the subjects of educational activities and their interaction.
- Realization of the teacher's ability to transfer knowledge to students, taking into account their mentality.
- Software implementation of the community of artificial agents with neural network architecture in the Java language in the MadKit tool environment.
Object of study : the system of teaching students in the department, which is heterogeneous and distributed. For modeling such systems, it is advisable to apply an agent-oriented approach.
Subject of research : the process of predicting residual knowledge of the student.
3. Review of research and development
Multi-agent systems are used in our life in graphic applications, for example, in computer games. Agent systems have also been used in films. The theory of MAS is used in composite defense systems. Also, MAS are used in transport, logistics, graphics, geographic information systems, robotics and many others. Multi-agent systems are well established in the field of networking and mobile technologies to ensure automatic and dynamic load balancing, extensibility and self-healing ability[2].
Complex agent-based systems have already found widespread use in industry. For example, IBM uses agents to produce semiconductor chips, Danish shipbuilding company – for welding holes in ships, and in Japan, the system based on agents performs the functions of the interface of the operator of superfast trains. MASs can be used both to design and simulate flexible production systems, and to manage real production systems (logistics), sales of products for various purposes (e-commerce), integration and knowledge management and scientific work[3].
.3.1 Overview of international sources
One of the most important works of the beginning of the 90s was the article by I. Shoem “Agent-oriented programming”. It described the social a look at the organization of calculations associated with the interaction of agents in the calculation process. In this case, the agent is considered as a “transparent box”. such “internal variables” (essentially, mental characteristics) as motives, beliefs, obligations, abilities to making and making decisions. The agent’s motives are the basis of his decisions, and beliefs determine the logical constraints on them. Agent communication carried out using communication protocols. In the context of the theory of speech acts, a standard set of primitives is introduced: “inform”, “Request”, “offer”, etc.[4].
In the USA, one of the first scientists who proposed to extend mental processes (or mental properties) to artificial objects, considered in the AI, and M.Minsky became interpret the mental sphere as a consequence of the interaction between active objects[5]. He described a number mechanisms for the emergence of intellectual behavior as a result of conflicts and cooperation between the simplest computing units, which he calls agents. Each of these agents is “responsible” for one or another mental property, and their interaction occurs spontaneously, without the involvement of any managing agent. Hence M.Minsky made an important conclusion that the functioning of the psyche is connected not so much with the implementation of a set of conclusions above character constructions, how many with the principles of self–organization in the interaction of autonomous processes[4].
3.2 Overview of national sources
In the 60s–70s in the USSR, it was proposed to describe the actions (behavioral acts) of autonomous agents by the action frames representing pairs of weighted graphs of a special type (“intention — realization”). In this model, such mental characteristics are already taken into account. and social factors as needs, motives, intentions, expected estimates, norms[4].
3.3 Overview of local sources
In Donetsk National Technical University (Department of Software Engineering) the study of multi-agent systems is carried out by my diploma supervisor, Ph.D., associate professor Fedyaev Oleg Ivanovich. For many years, he and his students have been considering aspects of agent-based modeling. Many theses and works at the conference are devoted to this topic.
Fedyaev O. along with past masters, he also did research in the area of multi-agent systems:
4. Agent-based modeling of the learning process of students at the level of competencies
Forecast subtasks
The task of predicting the quality of vocational training of students, depending on their personal characteristics and other factors is solved on the basis of the use of intelligent agents using artificial neural networks. To do this in the architecture of the agents of teachers appropriate artificial neural network models are introduced that are able to functionally describe the dependence of professional knowledge and skills in one discipline from factors affecting the completeness of this knowledge. In turn, the forecast task is divided into two subtasks[6].
Subtask 1. Setting up the model according to the observations. This is the inverse problem associated with finding the model parameters, i.e. with the construction of the function f according to the observed data of the student’s mentality, the teacher’s mentality, the learning environment, and the student’s knowledge and skills for the one studied discipline.
where MS is the mentality of the student; Tm – teacher mentality; E – learning environment; Рс – student's professionalism by one studied discipline.
Subtask 2. Formation of knowledge and skills on the mentality of the participants of the educational process. This task is to explicitly find student's professionalism (PS), i.e. his residual knowledge and skills, after studying a particular discipline, according to measured data on mentality student (MS) and teacher (Tm) using the built model f:
Gaia Methodology
The object of analysis is the system of teaching students in the department, which is heterogeneous and distributed. To simulate such systems, it is advisable to apply an agent-oriented approach. Agent-oriented analysis of the object, as a rule, is performed by Gaia Methodology[7].
It supports two levels of development of multi-agent systems: the micro level (development of a separate agent) and the macro level (development of an agency). It also has a significant limitation: the structure of each agent during its operation must remain unchanged. Gaia's methodology is from two big stages: analysis and design. At the stage of analysis, role models and interaction models are implemented, at the design stage - agent model (based on role models), service models and communication model (they are formed taking into account the models of the analysis stage). Stages Gaia's methodologies are presented in Figure 1.
Figure 1 – Gaia Methodology
Neural network means of intelligence agent implementation
The process of teaching students is to transfer knowledge and skills from teachers to students. The quality of training is fixed in examination record. The developed model of the learning process should form the output of residual knowledge of students on a separate discipline.
The forecast of residual knowledge for one specifically taken discipline for one student is carried out in two stages. At the first stage it is predicted examination score, and at the second stage, based on the predicted score, an average set of residual knowledge and skills is formed, corresponding to this assessment. The first neural network is trained on the basis of mental portraits of students and the examination sheet. The second neural network is based on the evaluation criteria and the curriculum of the discipline, which contains a list of knowledge and skills. The scheme of such a two-stage model is presented in Figure 2.
Figure 2 – A neural model diagram describing student learning outcomes using the example of one discipline
(animation: 9 frames, endlessly repeated cycles, 62 kilobytes)
Description of the structure of software agents in the MadKit environment
The MadKit tooling system that was used in the development is a modular and scalable multi-agent platform, written in java. It allows you to create software agents in different languages: Java, Python, Jess, Scheme, BeanSchell[8]. Presence of visual agent models increases the convenience of their design.
The MadKit architecture includes a graphical application host, application and system agents, and a system microkernel. The components of the microkernel of the system are the group / role manager, the synchronization mechanism and the messaging mechanism.
MadKit does not impose any restrictions on the architecture of agents to achieve maximum universality of applications. Agent Interaction implemented using asynchronous messaging. In a simulation model, an agent can send a message to another agent, defined by his address or by means of the broadcast message which is transferred to the agents playing this role in a certain group. Each agent have their own “mailbox” in which messages are delivered and which is checked by the agent for analyzing messages[9]. The program code of one of the mandatory sections of the student agent is presented in Figure 3.
Figure 3 – The program code of the agent section of the student
Findings
The technology itself of developing a multi-agent system is, of course, not simple, but not inaccessible to understanding. Development begins with the formation of the project, after which an analysis of the subject area is made. The results of the analysis are used to develop ontology specifications. After that, the agency architecture is selected based on the ontology specification. Then based on all the same specification of the ontology and the choice of the architecture of the agency produced by the specification of the behavior of the agent. The next step is to develop user interface libraries, for which the agent behavior specification is used, as well as the ontology specification. Besides, An agent behavior specification is used to develop an agent action library and agent specification. In turn, the library of interfaces user agent action library and agent specification are used to develop an agent program that, together with the period module execution allows you to create an agent application that is the basis for developing an intelligent agent application [10].
Prediction of learning outcomes will allow to analyze the quality of learning, to see the degree of student learning educational material, to detect the discrepancy in competencies between discipline and the requirements of firms, to assess the possibility of employment graduates. Analysis of the literature shows that the solution of these problems is carried out, as a rule, not by formal methods this reduces their practical value. The modern level of information technology allows us to develop qualitatively new models, combining the advantages of mathematical methods, statistics, the theory of neural networks, programming. With the advent of multi-agent theory systems have the opportunity to create models of complex distributed and inhomogeneous systems, the class of which includes the object under study [11].
List of sources
- Краткая история развития многоагентных систем [Электронный ресурс]. – Режим доступа: https://studopedia.ru/2_27513_kratkaya-istoriya-razvitiya-mnogoagentnih-sistem.html.
- Многоагентная система [Электронный ресурс]. – Режим доступа: https://ru.wikipedia.org/wiki/Многоагентная_система.
- Управление на базе мультиагентных систем [Электронный ресурс]. – Режим доступа: https://www.intuit.ru/studies/courses/13833/1230/lecture/24081.
- Агентно-ориентированного подхода [Электронный ресурс]. – Режим доступа: https://lektsia.com/1x885f.html.
- Minsky M. The Society of Mind. – NewYork: Simon and Shuster, 1986 [Электронный ресурс] – Режим доступа: http://www.acad.bg/ebook/ml/Society%20of%20Mind.pdf
- Лукина Ю.Ю., Федяев О.И. Технология создания мультиагентных систем в инструментальной среде MadKit/ Информационные управляющие системы и компьютерный мониторинг-2011 / Матеріали науково-технічної конференції молодих учених та студентів. - Донецьк, ДонНТУ - 2011 [Электронный ресурс] – Режим доступа: http://ea.donntu.ru/handle/123456789/12676
- Методология проектирования GAIA [Электронный ресурс] – Режим доступа: https://studopedia.ru/9_168381_metodologiya-proektirovaniya-Gaia.html.
- MadKit Development Giude [Электронный ресурс] – Режим доступа: http://www.madkit.net/documentation/devguide/devguide.html.
- Платформы для разработки МАС [Электронный ресурс] – Режим доступа: https://www.intuit.ru/studies/courses/13858/1255/lecture/23977.
- Иванов К.К., Лужин В.М., Кожевников Д.В. Программные агенты и мультиагентные системы // Молодой ученый. – 2017. – №7. – С. 11–13. https://moluch.ru/archive/141/39879/.
- Янкивский А.А., Павлова Е.М., Федяев О.И. Проектирование в среде MadKit агентно-ориентированной системы прогнозирования результатов обучения студентов/Программная инженерия: методы и технологии разработки информационно-вычислительных систем (ПИИВС-2018): сборник материалов II Международной научно-практической конференции (студенческая секция). 14–15 ноября 2018 г. – Донецк, ГОУ ВПО «Донецкий национальный технический университет», 2018. – с. 127–133.