Search

In this research, soccer task is investigated among the numerous tasks of deep reinforcement learning. The soccer task requires cooperative behavior. However, it is difficult for the agents to acquire the behavior, because a reward is sparsely given. Moreover, the behaviors of the allies and opponents must be considered by the agents. In addition, in the soccer task, if the agents attempt to acquire high-level cooperative behavior from low-level movements, such as ball kicking, a huge amount of time will be needed to learn a model. In this research, we conduct experiments in which reward shaping is incorporated into deep reinforcement learning. This enables the agents to efficiently acquire cooperative behavior from low-level movements in a soccer task. The findings of this research indicate that reward shaping with a designer's domain knowledge positively influences the agent's attempt to acquire cooperative behavior from low-level movements.

Research on generating natural language captions to visual data such as images and videos has produced considerable results with deep learning methods and attracted attention in recent years. In this research, we aim to generate recipe sentences from cooking videos acquired from YouTube. We treat the task as image captioning. There are two aspects to be considered in order to do so. We believe that the semantics of each process should be taken into account to improve the captioning ' s accuracy. Furthermore, data processing, that is obtaining images from each process using several visual processing methods such as object detection should be important. We propose a captioning model where a sentence vector is embedded to consider the consistency of the recipe. From differences between generated recipes and the reference recipe, we can calculate recipe scores. We use three metrics that are used in previous studies to evaluate the image captioning model. We compare the scores to with ones from baseline models.

Contemporary societies face many urban issues. To address these issues, governments, corporations and individuals should disclose and share their related statistical and sensory data. However, existing published data appear in various formats and contain defects. Therefore, few problems have been solved using these data. In this research, we sought to address this problem, by considering a university campus as a microcosm of society, designed data integration schema, and consolidated data into a knowledge graph. We then, applied and modified existing completion methods. In particular, regarding the bicycle environment, we trained our knowledge graph and evaluated it with the conventional method and our proposed derivative method, respectively. Using approximately 650 parking data with various dates and times, our method correctly estimated 54.5 more bicycles than the conventional method by comparing each time's mean absolute error.

Process aware information system (PAIS) is important in the recent business environment. Developments of PAIS need to consider contexts about technical and business elements. They are needed to develop PAIS effectively (e.g. monitoring environment and constructing adequate business process). Process mining is an important method for analyzing a business environment and utilizing PAIS development and improvement. LTL checking is an important method for checking a specific property to be satisfied with business processes, but correctly writing formal language like LTL is difficult. In this paper, we use LTL checking and prediction based on decision-tree learning for checking goal achievement, false detection and oversight detection. It helps writing properly LTL formula for representing the correct goal property. We conducted a case study using a real life log of traffic fine management process in Italy.

It is important to make sure that software satisfies stakeholders' requirements. However, as software has been getting more large-scale and complicated in recent years, it has certainly become more difficult to satisfy requirements. Even if there are goal-oriented requirements analysis approaches as techniques to systematically satisfy stakeholders' requirements, it is still difficult to maintain consistency of goal models and other requirements artifacts. In this paper, we propose a transformation approach that transforms models of KAOS, a well-known goal modeling methodology, into preliminary BPMN models by using refinement pattern of KAOS in a systematic way. It can assure consistency between systematically defined user's requirements and their realization process.

Recently, urban greening and agriculture have been receiving increased attention, but the cultivation of greenery is not a simple matter in a restricted space. Therefore, we propose Flower Voice, a voice assistant for smartphones that answers questions users will be faced on site, and provides a mechanism for registering the work. The system uses Linked Open Data as a knowledge source, and features improvemnt of accuracy based on user feedback and acquisitions of new data by user participation. This paper presents Plant LOD and an architecture of the sytem, and then evaluates the improvement of the accuracy.

We study a heterogeneous autonomous robot team composed of UAVs (Unmanned Air Vehicles) and UGVs (Unmanned Ground Vehicles) cooperate with each other to carry out global tasks. We introduce the system that UAVs and UGVs remove obstacles cooperatively. In our system, there are multiple type UGVs that have good points and shortcomings, and these UGVs remove a obstacle that block the way. Therefore, our systems can deal with more value of obstacles than existing systems. However our system is more complicated and has many problems to be solved. Therefore, we clarified the problems that our system need to be realized and implemented the robots. The system determines some ways to remove the obstacles using the aerial photograph taken by the UAV and the information of the obstacles by the UGVs, and then outputs a processed aerial photograph and some ways to remove the obstacles. The experimental results reminded us some necessary things for to carry out our goal. Our study has the goal that the autonomous robot team explores under unknown environment such as planetary explorations, or performs search and rescue operations in disaster areas.

Semantic search is intrinsically suited for information retrieval in field research, where keyword search is burdensome for users. Also, exploitation of mobile and facility sensors is now prevailing, but applications are still vague. Therefore, LOD can serve as an intermediary interpreting the semantics of the users and their environmental information obtained by the sensors and connecting it to the collective intelligence on the net. In this paper, to build ecosystem of LOD used in the field, we propose a mechanism of LOD content generation and its field application. Firstly, we introduce an Android application, which searches a plant fitting the environmental conditions obtained by the sensors from Plant LOD. Then, we show that our LOD generation mechanism builds the Plant LOD from the Web with an accuracy of 85 to 97%.

Consistency between various phases of the software development process is an important aspect in software engineering. Without this aspect, software products may not satisfy users' needs. The cause of the inconsistency is that users' requirements may not be sufficiently reflected in software products. We consider this situation as a gap between the requirements and the products. We focus on the early phases of the software development, because we need to deal with the problem before the inconsistency in the early phases affects the later phases. KAOS has been attracting attentions and UML has been widely adopted, in the early phases, for requirements analysis and design respectively. However, the requirements analysis phase and the design phase are often practiced separately which may cause a gap between them. In this paper, we propose a systematic approach, to reduce the gap between the requirement analysis phase and the design phase, by generating UML class diagram and OCL constraints from the KAOS model. We propose transformation rules between operation models and class diagrams and rules for producing OCL constraints from the KAOS model. In addition, we carried out a case study to confirm the validity of this approach.

Self-adaptive systems have recently been receiving much attention because of their ability to cope with the changes of environment, failures, and unanticipated events. These systems need an adaptation mechanism, which automatically computes the possible configurations, and decides the most appropriate configuration to fit the environment. In particular, the satisfaction of non-functional requirements must be considered when selecting the best reconfiguration. However, there are trade-off problems among non-functional requirements. Moreover, the adaptation mechanisms are typically developed separately from the components to be implemented, and it complicates the construction of such systems. We propose (1) a feature-oriented analysis technique, which can identify adaptation points, and calculate the contribution to non-functional goals of the configuration; (2) a component specification model, which extends an architectural description language for self-adaptation; (3) a reconfiguration framework aimed to reduce the complexity of the reconfiguration and generate the best configuration at run-time. We evaluate the feasibility of our framework by four different scenarios, and show that our framework reduces the complexity of the reconfiguration, and solves the trade-off problem among non-functional requirements.

Self-adaptive systems have recently attracted attention because of their ability to cope with changing environments, including system intrusions or faults. Such software must modify itself to better fit its environment, and one of the approaches by which we expect this capability to be achieved is the introduction of multiple control loops to assess the situation and to determine whether a change in behaviors or configurations is necessary and how to implement the change. Development of such systems with multiple control loops complicates the task of identifying components, and could be greatly aided by appropriate tool support. In this paper, we propose an architectural compiler for self-adaptive systems, which generates architectural configurations from the goal-oriented requirements descriptions. We also present a framework for generating such configurations with this compiler and a pattern in the requirements description. We evaluate the framework experimentally and show that it helps to generate suitable configurations that have high performance, and that the compiler scales well to large input models.

The purpose of this study is to reduce the difficulties encountered when designing multi-agent systems (MAS). Although MAS are one of the more effective approaches for dealing with the recent increase in software complexity, they are themselves difficult to develop. We believe the difficulties in determining agent responsibilities and the organizational structure as the most significant obstacles to MAS development. In this paper, we propose a design framework for MAS, which includes restriction rules in the goal-oriented requirements description and a generator that transforms the restricted requirements description into a general MAS design model. We demonstrate the effectiveness of our framework through a case study and show how the framework can be used to construct MAS design models more precisely than conventional methods permit.

In the web research, ontologies are widely used these days. However, several difficulties in ontology building have been pointed out so far and its costs are raising problems currently. In this paper, we propose ONTOMO that enables Internet users to take part in building ontologies as a part of collective intelligence. After introducing an overview of ONTOMO, an instance and property recommendation mechanism based on the editing history of multiple users will be presented together with experimental evaluations.

Many different kinds of mobile agent platforms have been developed. However, migration to a different type of mobile agent platform is actually impossible. As a solution to this problem, we propose an interoperability concept using incarnation agents. This concept realizes logical mobility between different kinds of agent platforms. The incarnation agent extracts an agent's procedures and status, compiles them into an agent platform-independent format, and then re-compiles them to the target agent platform format, thus enabling the process to continue. The incarnation agent also has autonomy for interoperability. It manages complex migration for interoperability so that the agent's procedures can be described in a simple mobile model. Moreover, the incarnation agent modifies the agent's procedures so that they are applicable to the facilities of the target agent platform. Feasibility experiments using interoperability middleware have been conducted as a basis for realizing incarnation agents.

Many people consider that security is one of the biggest problems for practical use of mobile agents that move around the network and do their tasks. In this paper, we assert that this issue can be effectively managed by using IPEditor, the development support tool of mobile multi-agent applications that we have been released, and Mobile UNITY, a formal specification language of mobile agent applications. IPEditor helps developers to design applications with visual supports of agent behaviors. In our method, we translate an IPEditor model to a Mobile UNITY program that is the formal specification of the agent behaviors. In addition, we describe the security requirements by the Mobile UNITY logic notation. Thus we can verify the security requirements by proving that the mobile UNITY program, therefore the IPEditor model, satisfies the mobile UNITY logic notation. We present an example of an electronic catalog (e-catalog) application and illustrate the effectiveness of our proposal.