EV3RT is, to our knowledge, the first RTOS-based software platform for LEGO Mindstorms EV3 robotics kit. It is faster and more suitable for developing applications with real-time requirements than other existing software platforms. In practice, EV3RT has been selected as one of the officially supported platforms of ET Robocon, a popular robot competition in Japan, since 2015 and used by many participating teams to make their robots accomplish the assigned tasks more stably and precisely. In this paper, the usage and architecture of EV3RT are firstly introduced. We then explain TOPPERS/HRP2 kernel, the RTOS of EV3RT, and how to use its protection functionalities to build a reliable platform. A mechanism to support dynamic module loading in a static RTOS is proposed to implement the application loader for EV3RT. Implementation techniques like approach to reusing Linux device drivers are also described. Finally, the advantages of EV3RT are shown by evaluating and comparing its performance with other platforms.
Supporting deaf and hard-of-hearing (D/HH) people is a crucial social welfare measure. However, currently, communication support for D/HH people is insufficient for a conversation with multiple hearing people. In order to support D/HH people in these situations, this paper proposes a multi-modal speech visualization application, named Mieta, which provides various aspects of information about speech content. For the purpose of evaluating how actually useful the Mieta is, we conduct an evaluation experiment with actual nine D/HH students. As a result, Mieta contributes significantly to understanding, although it increases misunderstanding for D/HH people when there are voice recognition errors. Furthermore, subjects find that the Mieta's strength lies in its main characteristics, which includes multimodal speech visualization, whereas some find weakness in its voice recognition performance speed.
The specification phase is an important phase in software development, and it is known that a defect in a specification seriously degrades the productivity and reliability of the overall development. Formal specification is a kind of functional modeling technique to define the functionality of the system rigorously at an appropriate level of abstraction using tools with mathematical backgrounds. The formal specification engineers perform the early stage of the modeling task, exploring the problem domain and learning the domain knowledge and requirements. This article explains the requirements on the support tools for the exploratory modeling, and introduces ViennaTalk, an integrated development environment for exploratory modeling in the formal specification language VDM-SL. Its design rationales, concrete design and implementation are also described.
In this paper, we propose a hybrid missing data technique combine deletion and imputation of missing data. First, the proposed method deletes projects and/or metrics with high missing rate. Next, the proposed method imputes using analogy based imputation for remaining missing data. In the experiment, we compared estimation accuracies of multivariate liner regression in software development dataset. The results showed that mean balanced relative error was improved than conventional method.
Multivariate regression models have been commonly used to estimate the software development effort to assist project planning and/or management. Since project data sets for model construction often contain missing values, we need to build a complete data set that has no missing values either by using imputation methods. However, while there are several ways to build the complete data set, it is unclear which method is the most suitable for the project data set. In this paper, using project data of 1364 cases (34% missing value rate) collected from several companies, we applied four imputation methods (k-nn method, applied CF method, Miss Forest method and Multiple Imputation method) to build regression models. Then, using project data of 160 cases (having no missing values), we evaluated the estimation performance of models after applying each imputation method. The result showed that Multiple Imputation method showed the best performance.
Efficient malware detection technologies include behavior-based methods using machine learning. In particular, online machine learning has remarkable advantages such as the capability of low-cost additional learning of new malware samples. The developers of a malware detection method employing online machine learning algorithms need to make an appropriate choice between the many algorithms that exist currently. However, there have been few comparative evaluations of multiple algorithms, and hence the knowledge needed to make a choice between them is lacking. In this study, we conducted a comparative evaluation of the characteristics of multiple online machine learning algorithms in behavior-based malware detection. We implemented four-category, seven-pattern algorithms (PA-I, PA-II, AROW, NAROW, NHERD, SCW-I, SCW-II), and quantitatively compared them by the standards of classification accuracy, learning speed, and classification speed. We assumed two use cases: initial screening of unknown program samples and real-time malware detection. We provided behavior logs of malware and benign applications to each algorithm, and performed learning and classification using it. Based on the result of the evaluation, we identified the best and worst algorithms in each use case and by each standard.