• Proceedings of the Korea Information Processing Society Conference
• /
• 2006.11a
• /
• pp.15-18
• /
• 2006

강화 학습은 시행착오를 통해 동적 환경과 상호작용하면서 학습을 수행하는 학습 방법으로 본 논문에서 테스트 환경으로 사용하는 Keepaway와 같은 동적인 환경에서 주로 사용하는 학습 방법이다. 본 논문에서는 학습을 통한 에이전트가 다른 정책을 사용하는 에이전트보다 성능이 더 높다는 것을 보이고자 한다. 학습 초기에는 다양한 전략을 평가하기 위해 최적이 아닌 행동을 선택하지만 시간이 지남에 따라 최적의 행동 선택에 수렴해 가는 것을 보이기 위한 실험을 수행한다. 이 실험을 통해 고정된 행동 양식을 가지는 정책보다 강화 학습을 이용한 에이전트들의 성능이 더 효과적인 것을 알 수 있었다.

• Journal of the Korea Society for Simulation
• /
• v.19 no.1
• /
• pp.153-160
• /
• 2010

Most part of errors in software development process are included during the stage of requirements definition and design. And correction or elimination of errors from those stages requires much more efforts and costs than those from the later part of software development process. However, despite of the importance of the validation of requirement definition and design stages, several kinds of problem have made it hard to be done successfully. Therefore, in this paper, we introduce a novel validation process for the preliminary design stage. The validation process is based on simulations of model and it can be used to validate requirements and model simultaneously. Models in the validation process will take only the behavior of software and be built on Ptolemy framework. The usability of our validation process is confirmed with a case study over DNS system environment. And the result of simulation shows well-known errors or vulnerabilities can be found with simulations of model which has the behavior of software. This means our validation process can be used as a process to validate requirements and models during the early stage of software development process.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.25 no.1
• /
• pp.22-28
• /
• 2015

In this paper, we studied the evolution of behavioral logic of artificial individuals using cell-level evolution framework. We first implemented cell-level evolution framework and then investigated the evolution of behavioral logic that artificial individuals ate foods on the framework. A logic frame for behavioral decisions of artificial individuals was devised and applied to the framework. From extensive tests, we found that most artificial individuals could evolve the behavioral logic that they could eat food in a short generation. It was also confirmed that most behavioral logics showed nearly same behaviors of artificial individuals in most tests. Our method has the differences from existing algorithms using evolutionary algorithms and evolvable hardwares in that it is a basically different approach. These results showed that our framework could be a good tool for investigating the evolution of artificial individuals in a cell-level.

• The KIPS Transactions:PartD
• /
• v.16D no.6
• /
• pp.921-926
• /
• 2009

Vulnerabilities in software can result in many social and economic problems once it has already been deployed and put to use. Thus, the vulnerabilities should be seriously taken into consideration from the beginning step of software development. A modeling and simulation method for software can be adopted as a testing tool for establishing vulnerability inspection strategies. For verification of usability of this strategy, in this paper, we modeled the behavior of a DNS system using Ptolemy and the simulation was performed. The result shows that a well-known vulnerability of DNS server could be effectively found, which confirms that the modeling and simulation can be used for vulnerability testing.