Volume 22 (2017): Issue 1 (December 2017)

The paper presents general guidelines for designing affective multi-agent systems (affective MASs). The guidelines aim at extending the existing agent-oriented software engineering (AOSE) methodologies to enable them to design affective MASs. The reason why affective mechanisms need specific attention during the design is the fact that the way how both rational tasks and interactions are done differ based on the affective state of the agents. Thus, the paper extends the traditional design approaches with the design of affective mechanisms and includes them in the design of the system as a whole.

Game-based learning as a learning approach has been popular for ages; however, game-based assessment as a trend started to evolve only few years ago. Since knowledge assessment is more associated with negative emotions, systems intended to assess knowledge should take into consideration emotions as well. The analysis of existing studies shows that systems with integrated game-based assessment seldom utilise learner’s emotions for provision of adaptation. The main aim of the present paper is to introduce an adaptation approach for to affective game-based assessment.

The paper explores geophysical methods of wells survey, as well as their role in the development of Kazakhstan’s uranium deposit mining efforts. An analysis of the existing methods for solving the problem of interpreting geophysical data using machine learning in petroleum geophysics is made. The requirements and possible applications of machine learning methods in regard to uranium deposits of Kazakhstan are formulated in the paper.

In the present paper, a new and improved visual sensor data fusion method is proposed that uses visible and far-infrared light sensors. Additionally, lux meter data are used for decision level fusion of beliefs of recognised target classes. The database consisting of 4 ambient light condition images is created using Canon and FLIR cameras.

The developed approach has been tested using database images, neural network training and classification, particularly for low light level conditions. Enhancements of target identification precision are proved by practical implementation and testing of the proposed method.

Model-Driven Software Development (MDSD) is a trend in Software Development that focuses on code generation from various kinds of models. To perform such a task, it is necessary to develop an algorithm that performs source model transformation into the target model, which ideally is an actual software code written in some kind of a programming language. However, at present a lot of methods focus on Unified Modelling Language (UML) diagram generation. The present paper describes a result of authors’ research on Two-Hemisphere Model (2HM) processing for easier code generation.

Implementation of an emulator of MIX, a mythical computer invented by Donald Knuth, is used as a case study of the features of the Scala programming language. The developed emulator provides rich opportunities for program debugging, such as tracking intermediate steps of program execution, an opportunity to run a program in the binary or the decimal mode of MIX, verification of correct synchronisation of input/output operations. Such Scala features as cross-compilation, family polymorphism and support for immutable data structures have proved to be useful for implementation of the emulator. The authors of the paper also propose some improvements to these features: flexible definition of family-polymorphic types, integration of family polymorphism with generics, establishing full equivalence between mutating operations on mutable data types and copy-and-modify operations on immutable data types. The emulator is free and open source software available at www.mix-emulator.org.

A technique of DropOut for preventing overfitting of convolutional neural networks for image classification is considered in the paper. The goal is to find a rule of rationally allocating DropOut layers of 0.5 rate to maximise performance. To achieve the goal, two common network architectures are used having either 4 or 5 convolutional layers. Benchmarking is fulfilled with CIFAR-10, EEACL26, and NORB datasets. Initially, series of all admissible versions for allocation of DropOut layers are generated. After the performance against the series is evaluated, normalized and averaged, the compromising rule is found. It consists in non-compactly inserting a few DropOut layers before the last convolutional layer. It is likely that the scheme with two or more DropOut layers fits networks of many convolutional layers for image classification problems with a plenty of features. Such a scheme shall also fit simple datasets prone to overfitting. In fact, the rule “prefers” a fewer number of DropOut layers. The exemplary gain of the rule application is roughly between 10 % and 50 %.