Summary on the final work
Contents
- Introduction
- 1. Topicality
- 2. The purpose and objectives of the study, expected results
- 3. A review of research and development
- 3.1 Overview of international sources
- 3.2 Overview of national sources
- 3.3 Overview of local sources
- 4. Neural network model of multi-agent system
- 4.1 Statement of the problem analysis preparing students
- 4.2 Neural network model based residual knowledge of students on their mentality
- Conclusions
- List of sources
Introduction
As the object of study is considered the university system of training professionals [1]. This system is composed of students, teachers and the learning environment is characterized by large inertia. From the standpoint of the theory of management in such systems the time period from the introduction of the perturbation (innovative changes in training) to obtain a response (output parameters characterizing parameters the quality of training) measured 4-6 years of training students in high school. Therefore, to apply techniques such as in-kind, and physical modeling [2–4] for research on effectiveness of the issuing Department of learning and teaching facilities is not possible. Moreover, if we consider that the educational process can not be described mathematically precisely because heterogeneity mnogoparametrichnosti, intelligence and complexity of the interaction of the structural elements, it remains the only way that can be applied to studies of this object – simulation. To build models object neural network methodology is appropriate to apply as a universal means of approximating a function that depends both on the amounts and on the quality of data.
Therefore, the aim of this work is to develop a neural network model that can describe the functional dependence of the students receive professional knowledge and skills of the factors affecting the completeness of the knowledge. To achieve this, you need to solve the following problems:
- to review the learning process of students of the University as a system with distributed intelligence;
- to identify internal and external factors affecting the quality of the assimilation of knowledge and skills;
- to develop a method for determining the mental and psycho-physiological characteristics of teachers and students;
- to develop the structure of neural network model and its algorithm for learning;
- to explore the complexity of setting up a model and value of its learning process (in the understanding of the learning process as a process of transfer of knowledge from teachers to students) [5].
1. Relevance of the topic
It is now one of the most promising approaches to the problem of modeling artificial intelligence, neural network training is multiagent approach. The relevance of the multi-agent simulation model is the ability to emulate the process trudoystroystva students to work. This in turn means analysis personal characteristics of the student, as well as modeling the interaction with the employer on the basis of the requirements for these techniques work [6].
A multi-agent system – a system formed by multiple interacting intelligent agents. Multi-agent systems can be used to solve such problems, It is difficult or impossible to solve with a single agent or a monolithic system. Examples of such problems are online trading, disaster response and modeling social structures.
2. The purpose and objectives of the study, deliverables
The aim of the study is to develop a multi-agent simulation model to predict the results of training and employment specialists
Research object: rarabotka and training neural networks.
Subject of research: agent-oriented development of neural network models to simulate the employment of students [7–10].
To achieve this, you need to solve the following problems:
- to review the learning process of students of the University as a system with distributed intelligence;
- to identify internal and external factors affecting the quality of the assimilation of knowledge and skills;
- to develop a method for determining the mental and psycho-physiological characteristics of teachers and students;
- to develop the structure of neural network model and its algorithm for learning;
- to explore the complexity of setting up a model and value of its learning process (in the understanding of the learning process as a process of transfer of knowledge from teachers to students).
3. Overview of Research and Development
Multi-agent systems based on neural network technology, which, in turn, is important and to apply the technology to this day. Data protection and their implementation are studied by scientists from different countries, their list includes Russia, China, USA, Germany, Japan and many others. Today, the main development of this technology is in Western countries, but the post-Soviet space and the list of Eastern countries gradually catch up with their Western counterparts studies [11–16].
3.1 Overview of international sources
Multi-agent systems are composed of several independent modules that handle objects with different knowledge and interests. This comprehensive section studies the agent systems, It is promising in the field of computer technology, because it covers not only the technique of neural network technology, but also uses the ideas of game theory, economics, operations research, logic, philosophy, linguistics, sociology, etc. [17]. It can serve as a tool for researchers in each of these areas, with proper use according to the methodology Hyulita K., D. Inman, Yoku Makoto et al. [8–12, 18]
While emphasizing the basics, the authors, such as John. Miller, D. Epstein, J.. Holland et al., Propose to use the volume and flow of knowledge to strict opredellennoy domain, as well as more carefully approach to distributed problem-solving game theory, multi-agent messaging and education agents social, design and modeling theory neyroalgoritmov distributed knowledge bases. These sources include: probability theory, classical logic, Markov model and programming principles of neural networks. [19–28]
According to D. O’Sullivan, B. Hernandez, D. Sellechu benefits of agent-based systems, as compared to other methods of modeling, can be expressed in three statements [29] agents define a phenomenon agent model provides a natural description of the system is flexible. Also, the agent-oriented systems able to withstand many adverse phenomena [30, 31], which implies other advantages of this approach [32].
Multi-agent systems are widely used in our lives. In addition to the basic methodologies for constructing neural network model multi-agent sitemy Samyuelson and Douglas The possibility of using agent-based models in the various spheres of life, including: graphics applications, have also been used in films, in composite systems Oboronservis, the transport sector, logistics, graphics, geographic information systems, the area of ??network and mobile technologies for provide automatic and dynamic balance of loading, scalability and resiliency [32–34].
M. Vulbridzh, Jose M. and others Cambridge University authors consider simulation as a whole, as well as how this essential agent-based modeling. In this context, simulation is a special case of mathematical modeling [35–39].
3.2 Overview of national sources
On. Subbotin means of study results in non-iterative, evolutionary and multi-agent model based synthesis methods based on neural networks. Considerable attention is paid analysis and classification of evolutionary methods, as well as multi-agent techniques. A wide range of new methods for modeling in problems of diagnosis, assessment, forecasting. Presents Examples of practical problems on the basis of the studied methods [14].
M Eigen and P. Schuster theory hypercycle- considered as one of the principles of self-organization of macromolecules. Presents research on the selection and evolution of RNA and DNA, produced a mathematical analysis of dynamical systems, to better understand the work of neural networks [15].
3.3 Overview of local sources
Though the subjects of development and training of neural network is a new agent, but in spite of that in the Donetsk National Technical actively being developed with respect to this area.
In particular article O. Fedjaeva devoted to the development of neural network model of the process of teaching students to agent system modeling of the labor market. The article discusses the development methodology neural network model that can describe the functional dependence of the students receive professional knowledge and skills of the factors affecting the completeness of the knowledge [1].
4. Neural network model of multi-agent system
4.1 Statement of the problem analysis preparing students
The problem of predicting the quality of vocational training of students according to their personal characteristics and other factors. She decided on the basis of application neural networks is to develop a neural network model that can describe the functional dependence of the students receive professional knowledge and skills in one discipline on factors affecting the completeness of the knowledge. In turn, it is divided into two sub-tasks.
1. Configuring Subtask models from observational data. This inverse problem associated with finding the model parameters, ie. E. With the construction of the function f of the observed data Mc, M, C and MS: where Ms – the mentality of the student; MP – the mentality of the teacher; C – learning environment; Pc – the professionalism of the student on a study discipline.
The mentality of the student (Ms) is determined by the elements that characterize its educational aspect and gained experience where m – the mentality; i – the intelligence; p – psychology; s – health.
The mentality of the teacher (M) in this case is determined by factors that affect the quality of the transmission of knowledge from teacher to student where us – academic degree; uz – academic status; h – experience; v – age; a – artistry.
The learning environment © is characterized by the state of training and methodological and technical support of the educational process, as well as the level of organization of student learning.
The professionalism of the student on a study discipline (Pc) is determined by the amount of knowledge (zc) and skills (uc), which it receives in the course of studying this discipline where Zd – the amount of knowledge, defined curriculum discipline, which is read at the department; Z – the amount of knowledge on the professional direction, defined the current state of science and technology;
Subtask 2. Formation of knowledge and skills on the mentality of participants in the educational process. This task consists in the explicit finding professional student (Pc), that. E. Its knowledge and skills after studying particular discipline, from the measured data on the mentality of the student (MS) and a teacher (M) with the help of the constructed model f.
This sub-task belongs to a class predictive tasks. You can use it to investigate the influence of various parameters (the content of the curriculum, the number of students, and so on. D.) The quality of education in particular the University.
4.2 Neural network model based residual knowledge of students on their mentality
In order to construct the model should take into account factors (personal characteristics) that affect the quality of students’ knowledge assimilation. We used the popular methods of psychological analysis [5].
In our opinion, the main factors affecting the absorption of student teaching material, can be systematized as shown in Fig. 1. Each of these is divided into several indicators which can be determined by the results of tests, interviews, and so on. d. [5, 6]. The analysis of these factors allows us to study the identity of the student with the different parties, to identify the most important mental particularly affecting the success of training.
Figure 1 – Factors affecting the absorption of the material the student
There have been developed methods for determining the mental and psycho-physiological characteristics of the student. Evaluation results of each of these parameters may be systematized and standardized. These techniques combine to form a system that determines the mental portrait of the student. Table. 1 illustrates this systematization.
Following the completion of surveys and tests will be determined by a multi-portrait student, which can be used to develop a model of knowledge transfer.
The learning process of students is to transfer the knowledge and skills of teachers. The quality of education is recorded in the examination sheet. The developed model of the process Training should form the output residual knowledge of the student on a separate discipline with which he goes to the labor market. According to them, employers decide on the employment of candidates for vacant positions.
The forecast residual knowledge of a specific discipline taken for one student in two stages. The first step is predicted evaluation examination, and the second stage, based on the predicted estimates formed a set of averaged residual knowledge and skills relevant to this assessment. The first neural network will be trained on the basis of mental portrait of a group of students and the examination sheet. Second neural network – based on the evaluation criteria and the curriculum of discipline, which contains a list of knowledge and skills.
The output signals of the second neural network to form a vector whose components are fixed presence or absence of the residual knowledge or skills. The size of the vector is determined by the total amount of knowledge and skills provided by the curriculum subjects. They are designated by the vector Y=(y1, y2, ..., yn), where n – number of knowledge and skills.
The structure of both neural networks belongs to a class of homogeneous multi-layer perceptrons with complete serial communications with the sigmoid activation function. Training neural networks was carried out by the strategy of «supervised learning» algorithm of back propagation. The training set for the second neural network is teacher professional (expert) in their discipline, using approved evaluation criteria and curriculum of discipline, which contains a list of knowledge and skills. To test the adequacy of the neural network model as a simulation environment of artificial neural networks used bag Neural Network Toolbox, which is included in the standard supply MATLAB. Driving a two-stage model is shown in Fig. 2.
Figure 2 – Neural model describing the results of the professional training of the student on the example of one discipline
(animation: 9 slides, 10 cicles, 21.3 kb)
When building a training set for the first neural network were selected 6 students listen to a training course «Intelligent Systems in the economy» and has received test scores. Students were selected for testing in such a way that in the training set were presented all test scores. These students were tested according to the method described in this section. For the training set data were taken first five students. The results of the student number 6 will be used for testing a trained neural network. The training set for a second neural network should prepare a teacher who reads the students educational discipline. From the curriculum (which is approved by the regulatory document) was taken a list of knowledge and skills, that the student must master in the discipline, and for him the teacher formed a table showing, for the knowledge and skills put a certain score.
The joint work of the two trained neural network was evaluated on the characteristics of the mentality of the student with the number 6, which did not participate in training. The simulation was performed in according to a two-stage circuit in Fig. 2. Analysis of the results of the first stage showed that the values ??of the components of the output vector are similar to the code (1,0,0,0). This encoding matches Examination assessment «unsatisfactory», which in reality was this student in the exam.
The forecasted estimate output from the first neural network was applied to a second input of the neural network, which has shaped the resulting vector Y residual knowledge and skills of students (Fig. 3).
Figure 3 – Forecasted residual knowledge and skills for the sixth student academic discipline Review
The values of the components of Y can be interpreted as the degree of confidence that this student are saved in its memory the relevant knowledge and skills (of course, with respect to use of training sets). If we compare this result with the evaluation criteria of educational discipline «Intelligent systems in the economy», that sets of knowledge and skills predicted corresponds to the evaluation «unsatisfactory».
Conclusions
An approach to the neural network modeling is difficult to formalize the process of vocational training of young professionals, based on a simulation of the transfer of skills and knowledge, depending on the personal characteristics of the students.
established internal and external factors affecting the performance of students and the quality of assimilation of knowledge and skills. Particular attention was paid to the student as an individual and his place in the learning process. Based on this, we developed a special technique that analyzes the psychological, emotional, environmental and physical abilities of the student.
After receiving mental portrait student developed a two-stage construction of neyroalgoritm neyromodeli simulating the result of vocational training by identifying residual knowledge and skills of the student, which will be used in the labor market.
The preliminary results of a study on software models showed the correctness of the proposed ideas for solving this problem.