An Efficient Topology/Parameter Control in Evolutionary Design for Multi-domain Engineering Systems

Kisung Seo 한국지능시스템학회 2005 INTERNATIONAL JOURNAL of FUZZY LOGIC and INTELLIGE Vol.5 No.2

This paper suggests a control method for an efficient topology/parameter evolution in a bond graph-based GP design framework that automatically synthesizes designs for multi-domain, lumped parameter dynamic systems. We adopt a hierarchical breeding control mechanism with fitness-level-dependent differences to obtain better balancing of topology/parameter search ? biased toward topological changes at low fitness levels, and toward parameter changes at high fitness levels. As a testbed for this approach in bond graph synthesis, an eigenvalue assignment problem, which is to find bond graph models exhibiting minimal distance errors from target sets of eigenvalues, was tested and showed improved performance for various sets of eigenvalues.

Tree-Structure-Aware Genetic Operators in Genetic Programming

Seo, Kisung,Pang, Chulhyuk The Korean Institute of Electrical Engineers 2014 Journal of Electrical Engineering & Technology Vol.9 No.2

In this paper, we suggest tree-structure-aware GP (Genetic Programming) operators that heed tree distributions in structure space and their possible structural difficulties. The main idea of the proposed GP operators is to place the generated offspring of crossover and/or mutation in a specified region of tree structure space insofar as possible by biasing the tree structures of the altered subtrees, taking into account the observation that most solutions are found in that region. To demonstrate the effectiveness of the proposed approach, experiments on the binomial-3 regression, multiplexor and even parity problems are performed. The results show that the results using the proposed tree-structure-aware operators are superior to the results of standard GP for all three test problems in both success rate and number of evaluations.

Evolutionary Design for Multi-domain Engineering system-Air Pump

Kisung Seo(서기성) 한국지능시스템학회 2005 한국지능시스템학회 학술발표 논문집 Vol.15 No.2

This paper introduces design method for air pump system using bond graph and genetic programming to maximize outflow subject to a constraint specifying maximum power consumption. The air pump system is a mixed domain system which includes electromagnetic, mechanical and pneumaticelements. Therefore an appropriate approach for a better system for synthesis is required. Bond graphs are domain independent, allow free composition, and are efficient for classification and analysis of models. Genetic programming is well recognized as a powerful tool for open-ended search. The combination of these two powerful methods for evolution of multi-domain system, BG/GP, was tested for redesign of air pump system.

A Comparative Study among Three Automatic Gait Generation Methods for Quadruped Robots

SEO, Kisung,HYUN, Soohwan 'Institute of Electronics, Information and Communi 2014 IEICE transactions on information and systems Vol.97e.d No.2

Tree-Structure-Aware Genetic Operators in Genetic Programming

Kisung Seo,Chulhyuk Pang 대한전기학회 2014 Journal of Electrical Engineering & Technology Vol.9 No.2

In this paper, we suggest tree-structure-aware GP (Genetic Programming) operators that heed tree distributions in structure space and their possible structural difficulties. The main idea of the proposed GP operators is to place the generated offspring of crossover and/or mutation in a specified region of tree structure space insofar as possible by biasing the tree structures of the altered subtrees, taking into account the observation that most solutions are found in that region. To demonstrate the effectiveness of the proposed approach, experiments on the binomial-3 regression, multiplexor and even parity problems are performed. The results show that the results using the proposed tree-structure-aware operators are superior to the results of standard GP for all three test problems in both success rate and number of evaluations.

Hierarchical Topology/Parameter Evolution in Engineering Design

Kisung Seo(서기성) 한국지능시스템학회 2005 한국지능시스템학회 학술발표 논문집 Vol.15 No.1

This paper suggests a control method for efficient topology/parameter evolution in a bond-graph-based GP design framework that automatically synthesizes designs for multi-domain, lumped parameter dynamic systems. We adopt a hierarchical breeding control mechanism with fitness-level-dependent differences to obtain better balancing of topology/parameter search - biased toward topological changes at low fitness levels, and toward parameter changes at high fitness levels. As a testbed for this approach, an eigenvalue assignment problem, which is to find bond graph models exhibiting minimal distance errors from target sets of eigenvalues, was tested and showed improved performance for various sets of eigenvalues.

Evolutionary Design for Multi-domain Engineering System - Air Pump Redesign

Kisung Seo 한국지능시스템학회 2006 한국지능시스템학회논문지 Vol.16 No.2

This paper introduces design method for air pump system using bond graph and genetic programming to maximize outflow subject to a constraint specifying maximum power consumption. The air pump system is a mixed domain system which includes electromagnetic, mechanical and pneumatic elements. Therefore an appropriate approach for a better system for synthesis is required. Bond graphs are domain independent, allow free composition, and are efficient for classification and analysis of models. Genetic programming is well recognized as a powerful tool for open-ended search. The combination of these two powerful methods, BG/GP, was tested for redesign of air pump system.

멀티-에너지 도메인 동적 시스템을 위한 본드그래프/유전프로그래밍 기반의 자동설계 방법론

서기성(Kisung Seo) 한국지능시스템학회 2006 한국지능시스템학회논문지 Vol.16 No.6

멀티-도메인 공학시스템은 전기, 기계, 유압, 열등의 구성요소를 포함하고, 시스템 구성이 복잡하여 설계에 많은 어려움을 가지고 있다. 최적의 설계를 위해서는 각 도메인에 대한 통합된 설계 방법과 자동적이고 효율적인 탐색방법이 요구된다. 본 논문은 도메인에 독립적인 모델링 도구인 본드 그래프(Bond Graph)와 대규모 공간 해의 탐색에 접합한 진화 알고리즘의 일종인 유전 프로그래밍(Genetic Programming)를 결합하여 멀티 도메인 동적시스템에 대한 디자인 해를 자동적으로 생성해주는 설계 방법을 제시하였다. 제안된 설계방법의 효용성을 입증하기 위해서 3 가지 서로 다른 도메인을 가진 아나로그 필터, 전동프린터 드라이브, 에어펌프 시스템에 대한 설계 결과가 기술된다. Multi-domain design is difficult because such systems tend to be complex and include a mixtures of, electrical, mechanical, hydraulic, and thermal components. To design an optimal system, unified and automated procedure with efficient search technique is required. This paper introduces design method for multi-domain system to obtain design solutions automatically, combining bond graph which is domain independent modeling tool and genetic programming which is well recognized as a powerful tool for open-ended search. The suggested design methodology has been applied for design of electric filter, electric printer drive, and air pump system as a proof of concept for this approach.

진화연산 기반 CNN 필터 축소

서기성(Kisung Seo) 대한전기학회 2018 전기학회논문지 Vol.67 No.12

A convolutional neural network (CNN), which is one of the deep learning models, has been very successful in a variety of computer vision tasks. Filters of a CNN are automatically generated, however, they can be further optimized since there exist the possibility of existing redundant and less important features. Therefore, the aim of this paper is a filter reduction to accelerate and compress CNN models. Evolutionary algorithms is adopted to remove the unnecessary filters in order to minimize the parameters of CNN networks while maintaining a good performance of classification. We demonstrate the proposed filter reduction methods performing experiments on CIFAR10 data based on the classification performance. The comparison for three approaches is analysed and the outlook for the potential next steps is suggested.

직교좌표공간과 관절공간에서의 4족 보행로봇의 두 가지 진화적 걸음새 생성기법

서기성(Kisung Seo) 대한전기학회 2014 전기학회논문지 Vol.63 No.3

Two evolutionary gait generation methods for Cartesian and Joint coordinates space are compared to develop a fast locomotion for quadruped robots. GA(Genetic Algorithm) based approaches seek to optimize a pre-selected set of parameters for the locus of paw and initial position in cartesian coordinates space. GP(Genetic Programming) based technique generate few joint trajectories using symbolic regression in joint coordinates space as a form of polynomials. Optimization for two proposed methods are executed using Webots simulation for the quadruped robot which is built by Bioloid. Furthermore, simulation results for two proposed methods are analysed in terms of different coordinate spaces.