• ETRI Journal
• /
• v.42 no.4
• /
• pp.505-517
• /
• 2020

The increasing size and complexity of deep neural networks (DNNs) necessitate the development of efficient high-performance accelerators. An efficient memory structure and operating scheme provide an intuitive solution for high-performance accelerators along with dataflow control. Furthermore, the processing of various neural networks (NNs) requires a flexible memory architecture, programmable control scheme, and automated optimizations. We first propose an efficient architecture with flexibility while operating at a high frequency despite the large memory and PE-array sizes. We then improve the efficiency and usability of our architecture by automating the optimization algorithm. The experimental results show that the architecture increases the data reuse; a diagonal write path improves the performance by 1.44× on average across a wide range of NNs. The automated optimizations significantly enhance the performance from 3.8× to 14.79× and further provide usability. Therefore, automating the optimization as well as designing an efficient architecture is critical to realizing high-performance DNN accelerators.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.36 no.3
• /
• pp.63-70
• /
• 2013

Many real world optimization problems are discrete and multi-valued. Meta heuristics including Genetic Algorithm and Particle Swarm Optimization have been effectively used to solve these multi-valued optimization problems. However, extensive comparative study on the performance of these algorithms is still required. In this study, performance of these algorithms is evaluated with multi-modal and multi-dimensional test functions. From the experimental results, it is shown that Discrete Particle Swarm Optimization (DPSO) provides better and more reliable solutions among the considered algorithms. Also, additional experiments shows that solution quality of DPSO is not lowered significantly when bit size representing a solution increases. It means that bit representation of multi-valued discrete numbers provides reliable solutions instead of becoming barrier to performance of DPSO.

• Proceedings of the Korean Society of Propulsion Engineers Conference
• /
• 2003.10a
• /
• pp.145-150
• /
• 2003

The present study aims to optimize the performance of the Two-Stage Gas Gun by using the experimentally obtained data. RSM(Response Surface Method) was adopted in the optimization process to find the operating parameter than can maximize the projectile speed with the minimum number of tests. To decide the test points which results can consist of the response surface, 3$^{k}$ full factorial method was used, and the design variables were chosen with piston mass and 2$^{nd}$ driver fill pressure. The response surface was composed by nine test results and consequently the optimization was done with GENOCOP III, inherently GA code, in order to seek the optimal test point. The optimal test condition from the response surface was verified by the experiment. Results showed that the optimization process with response surface can successfully predict the test results fairly well. This study shows the possibility of performance optimization for the experimental facilities using numerical optimization algorithm.

• Journal of Mechanical Science and Technology
• /
• v.18 no.10
• /
• pp.1829-1836
• /
• 2004

This study presents the performance optimization of hypervelocity launcher system by using the experimentall data. During the optimization, the RSM (Response Surface Method) is adopted to find the operating parameters that could maximize the projectile speed. To construct a reliable response surface model, 3 full factorial method is used with the selected design variables, such as piston mass and 2 driver fill pressure. Nine test data could successfully construct the reasonable response surface, which used to yield the optimal operational conditions of the system using the genetic algorithm. The optimization results are confirmed by the experimental test with a good accuracy. Thus, the optimization can improve the performance of the facility.

• Advances in Computational Design
• /
• v.2 no.3
• /
• pp.241-256
• /
• 2017

This paper introduces the process for Multi-objective Optimization Framework (MOF) which mediates multiple conflicting design targets. Even though the extensive researches have shown the benefits of optimization in engineering and design disciplines, most optimizations have been limited to the performance-related targets or the single-objective optimization which seek optimum solution within one design parameter. In design practice, however, designers should consider the multiple parameters whose resultant purposes are conflicting. The MOF is a BIM-integrated and simulation-based parametric workflow capable of optimizing the configuration of building components by using performance and non-performance driven measure to satisfy requirements including build programs, climate-based daylighting, occupant's experience, construction cost and etc. The MOF will generate, evaluate all different possible configurations within the predefined each parameter, present the most optimized set of solution, and then feed BIM environment to minimize data loss across software platform. This paper illustrates how Multi-objective optimization methodology can be utilized in design practice by integrating advanced simulation, optimization algorithm and BIM.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.29 no.4
• /
• pp.117-134
• /
• 2004

This study suggests a data driven optimization approach, which simulates the models of human learning processes from cognitive sciences. It shows how the human learning processes can be simulated and applied to solving combinatorial optimization problems. The main advantage of using this method is in applying it into problems, which are very difficult to simulate. 'Undecidable' problems are considered as best possible application areas for this suggested approach. The concept of an 'undecidable' problem is redefined. The learning models in human learning and decision-making related to combinatorial optimization in cognitive and neural sciences are designed, simulated, and implemented to solve an optimization problem. We call this approach 'SLO : simulated learning for optimization.' Two different versions of SLO have been designed: SLO with position & link matrix, and SLO with decomposition algorithm. The methods are tested for traveling salespersons problems to show how these approaches derive new solution empirically. The tests show that simulated learning for optimization produces new solutions with better performance empirically. Its performance, compared to other hill-climbing type methods, is relatively good.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.3
• /
• pp.432-436
• /
• 2009

We have developed a new optimization algorithm termed simple bacteria cooperative optimization (sBCO) based on bacteria behavior patterns [1]. In [1], we have introduced the algorithm with basic operations and showed its feasibility with some function optimization problems. Since the sBCO was the first version with only basic operations, its performance was not so good. In this paper, we adopt a new operation, rank replacement, to the sBCO for improving its performance and compare its results to those of the simple genetic algorithm (sGA) which has been well known and widely used as an optimization algorithm. It was found from the experiments with four function optimization problems that the sBCO with rank replacement was superior to the sGA. This shows that our algorithm can be a good optimization algorithm.

• Journal of Digital Convergence
• /
• v.10 no.4
• /
• pp.207-215
• /
• 2012

This paper proposes a proxy-based caching optimization scheme for improving the streaming media services in wireless mobile ad hoc networks. The proposed scheme utilizes the proxy for data packet transmission between media server and nodes in WLANs, and the proxy locates near the wireless access pointer. For caching optimization, this paper proposes NFCO (non-full cache optimization) and CFO (cache full optimization) scheme. When performs the streaming in the proxy, the NFCO and CFO is to optimize the caching performance. This paper compared the performance for optimization between the proposed scheme and the server-based scheme and rate-distortion scheme. Simulation results show that the proposed scheme has better performance than the existing server-only scheme and rate distortion scheme.

• Transactions of the Korean Society of Automotive Engineers
• /
• v.19 no.4
• /
• pp.99-106
• /
• 2011

This paper deals with engine room layout design optimization of fuel cell electric vehicle (FCEV), which has been proposed as a potential alternative to fossil fuel depletion. Investing the great R&D efforts, the global vehicle manufacturers, especially Honda motor corporate, have shown not prototype vehicle but commercial vehicle using fuel cell in the market recently. In this paper, we analyze cooling performance and flow characteristic in the engine room of newly FCEV, in addition we suggest the optimization process for engine room layout design optimization. The two radiators in the vehicle for fuel cell stack and electronic components cooling have been analyzed and their performance are obtained in terms of cooling performance ratio (CPR). The value of CPR should always be less than one and based on criteria, we have achieved the optimum cooling performance of radiators for stack and electronic components. Aerodynamic performance is evaluated in terms of drag coefficient, improved through underbody modification using air devices.

• Transactions of the Korean Society of Automotive Engineers
• /
• v.16 no.1
• /
• pp.175-180
• /
• 2008

Due to the global environment problems and the consumer's need for higher vehicle performance, it becomes very important for the global car makers to reduce vehicle weight. To reduce vehicle weight, many car makers have tried to use lightweight materials, for example, aluminum, magnesium, and plastics, for the vehicle structures and components. Especially, the ASF(aluminum space frame) is known for the excellent concept of the vehicle to satisfy structural rigidity, safety performance and weight reduction. In this research, the design of experiments and the multi-disciplinary optimization technique were utilized to meet the weight and structural rigidity target of the ASF. For the structural performance of the ASF, the locations and the size of aluminum extruded frames, aluminum cast nodes, and the aluminum sheets were optimized. As a result, the optimization design procedure has been set up to meet both structural and weight target of the ASF, and the assembled ASF showed good structural performance and weight reduction.