Genetic Algorithm (GA), which is widely known as a general-purpose optimization method based on genetic evolution, has essential difficulties in its huge computation time and premature convergence. In order to overcome these difficulties and to search a new application, we propose a dedicated processor architecture, which can provide high-speed and high-expandable GA processing using VLSI multi-processor approach based on Distributed GA. A VLSI implementation of a processor element (PE), which is characterized by parallel evolutionary pipelines and adaptive genetic operations, indicates that the PE can be 130 times faster than conventional software processing. Furthermore, the parallel computer simulation demonstrates that the GA processor, with a newly proposed hierarchical ring topology, can provide a scalable performance according to PE numbers and a potential capability for real-time GA processing.

View full abstract

In this paper, we apply the minimum variance estimation algorithm using a Kalman filter to the waveform data of auditory brainstem response (ABR). The model parameters that extract the feature are obtained effectively by our proposed method. The ABR is usually analyzed by the batch processing through addition for a few minutes because of its slight potential. We will expand the possibility of measuring it online by minimizing the model error based on a Kalman filter. We especially discuss the type and the degree of the estimated transfer function. We use the ABR waves obtained by averaging (traditional method) less than 2000 times and estimate its transfer function to extract the feature using a Kalman filter. The result of our method showed a higher correlation than traditional method of averaging to the ABR signal. If it is possible to apply this method to clinical cases, we can expect much reduction of the addition of signals which contributes to speed-up the monitor diagnoses in the emergency medical care and the operation, etc.

View full abstract

In this paper, a new algorithm for discrete optimization problems is proposed. The algorithm is called “Random Search Method with Intensification and Diversification - Discrete Version ‘RasID-D’. In intensification phase, RasID-D searches the neighborhood with small range, while in diversification phase, searches the large neighborhood. Therefore, RasID-D can search good solutions rapidly and can escape from local minimum. Simulation studies show that RasID-D is more useful and effective than the conventional methods for a certain discrete optimization problems.

View full abstract