• The Korean Journal of Applied Statistics
• /
• v.26 no.5
• /
• pp.847-856
• /
• 2013

We propose a modified learning process for generalized neural network using a learning algorithm by Liu et al. (2001). We consider the effect of initial weights, training results and learning errors using a modified learning process. We employ an incremental training procedure where training patterns are learned systematically. Our algorithm starts with a single training pattern and a single hidden layer neuron. During the course of neural network training, we try to escape from the local minimum by using a weight scaling technique. We allow the network to grow by adding a hidden layer neuron only after several consecutive failed attempts to escape from a local minimum. Our optimization procedure tends to make the network reach the error tolerance with no or little training after the addition of a hidden layer neuron. Simulation results with suitable initial weights indicate that the present constructive algorithm can obtain neural networks very close to minimal structures and that convergence to a solution in neural network training can be guaranteed. We tested these algorithms extensively with small training sets.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2005.06a
• /
• pp.1883-1886
• /
• 2005

In this paper, a decoupled dual servo (DDS) stage for ultra-precision scanning system with large working range is introduced. In general, dual servo systems consist of a fine stage for short range and a coarse stage for long range. The proposed DDS also consists of a $XY\theta$ fine stage for handling and carrying workpieces and one axis coarse stage. Its coarse stage consists of air bearing guide system and a coreless linear motor with force ripple. The fine has four voice coil motors(VCM) as its actuator. According to a VCM's nature, there are no mechanical connections between coils and magnetic circuits. Moreover, VCM doesn't have force ripples due to imperfections of commutation components of linear motor systems - currents and flux densities. However, due to the VCM's mechanical constraints the working range of the fine is about $25mm^2$. To break that hurdle, the coarse stage with linear motors is used to move the fine about 500mm. Because of the above reasons, the proposed DDS can achieve higher precision scanning than other stages with only one servo. With MATLAB's Sequential Quadratic Programming (SQP), the VCMs are optimally designed for the highest force under conditions and constraints such as thermal dissipations due to its coil, its size, and so on. And for their movements without any frictions, guide systems of the DDS are composed of air bearings. To get precisely their positions, a linear scale with 5nm resolution are used for the coarse stage's motion and three plane mirror laser interferometers with 5nm for the fine's $XY\theta$ motions. With them, on scanning the two stages have same trajectories. The control algorithm is named Parallel method. The embodied ultra-precision scanning system has sub 100nm following error and in-positioning stability.

• The KIPS Transactions:PartB
• /
• v.9B no.1
• /
• pp.1-6
• /
• 2002

Many customers consult the Internet before making purchase goods and using contents. The systems in the Internet could store a lot of data and classify the data into information to get relationship between a company and customers. To do that, let's consider a knowledge-assisted hybrid system that utilizes individually a customer's preference to make an optimal solution in the his/her decision making. The knowledge made by using the preference is employed to select an domain set appropriate to him/her business, and the process of selecting definitely provides the customer some benefits: elimination of discomfort from unknown information and reduction of costs and search time for forming an suitable domain set. To effectively adopt individual customer's preference and actively adapt change of business situation, this study propose an architecture of the system which includes rule presentations and an inference engine, and integrates a knowledge-based component into a quadratic programming component. In the experimental results, it is found that a knowledge-assisted hybrid system implemented by this idea is more flexible than existing systems in extension of knowledge about an customer's preference and goes beyond the traditional models.