• The Transactions of the Korea Information Processing Society
• /
• v.3 no.2
• /
• pp.381-389
• /
• 1996

A simple modification of the standard back-propagation algorithm to eliminate redundant connections(weights and biases) is described. It was motivated by speculations from the distribution of the magnitudes of the weights and the biases, analysis of the classification boundary, and the nonlinearity of the sigmoid function. After initial training, this algorithm eliminates all connections of which magnitude is below a threshold by setting them to zero. The algorithm then conducts retraining in which all weights and biases are adjusted to allow important ones to recover. In studies with Boolean functions, the algorithm reconstructed the theoretical minimum architecture and eliminated the connections which are not necessary to solve the functions. For simulated random signal classification problems, the algorithm produced the result which is consistent with the idea that easier problems require simpler networks and yield lower misclassification rates. Furthermore, in comparison, our algorithm produced better generalization than the standard algorithm by reducing over fitting and pattern memorization problems.

• Journal of Korea Society of Industrial Information Systems
• /
• v.11 no.3
• /
• pp.20-33
• /
• 2006

Because of rapid expansion of third party logistics, fierce competition in the transportation industry, and the diversification and globalization of transportation channels, an effective transportation planning by means of multimodal transport is badly needed. Accordingly, this study aims to suggest an optimal transport algorithm for the multimodal transport in the international logistics. As a solution for this problem, first of all, we have applied a pruning algorithm to simplify it, suggesting a heuristic algorithm for constrained shortest path problem to find out a feasible area with an effective time range, which has been applied to the Label Setting Algorithm, consequently leading to multiple Pareto optimal solutions. Also, in order to test the efficiency of the algorithm for constrained shortest path problem, this paper has applied it to the actual transportation path from Busan port of Korea to Rotterdam port of Netherlands.

• Journal of KIISE
• /
• v.43 no.2
• /
• pp.246-255
• /
• 2016

The orthogonal frequency-division multiple access (OFDMA) systems are well suited to multi-rate multicast transmission, as they allow flexible resource allocation across both frequency and time, and provide adaptive modulation and coding schemes. Unlike layered video coding, the multiple description coding (MDC) enables flexible decomposition of the raw video stream into two or more substreams. The quality of the video stream is expected to be roughly proportional to data rate sustained by the receiver. This paper describes a mathematical model of resource allocation and throughput in the multi-rate video multicast for the OFDMA wireless and mobile networks. The impact on mean opinion score (MOS), as a measurement of user-perceived quality (by employing a variety of scheduling disciplines) is discussed in terms of utility maximization and proportional fairness. We propose a pruning algorithm to ensure a minimum video quality even for a subset of users at the resource limitation, and show the optimal number of substreams and their rates can sustain.

• Journal of Computing Science and Engineering
• /
• v.1 no.2
• /
• pp.177-194
• /
• 2007

Index selection is one of the most important decisions to take in the physical design of relational data warehouses. Indices reduce significantly the cost of processing complex OLAP queries, but require storage cost and induce maintenance overhead. Two main types of indices are available: mono-attribute indices (e.g., B-tree, bitmap, hash, etc.) and multi-attribute indices (join indices, bitmap join indices). To optimize star join queries characterized by joins between a large fact table and multiple dimension tables and selections on dimension tables, bitmap join indices are well adapted. They require less storage cost due to their binary representation. However, selecting these indices is a difficult task due to the exponential number of candidate attributes to be indexed. Most of approaches for index selection follow two main steps: (1) pruning the search space (i.e., reducing the number of candidate attributes) and (2) selecting indices using the pruned search space. In this paper, we first propose a data mining driven approach to prune the search space of bitmap join index selection problem. As opposed to an existing our technique that only uses frequency of attributes in queries as a pruning metric, our technique uses not only frequencies, but also other parameters such as the size of dimension tables involved in the indexing process, size of each dimension tuple, and page size on disk. We then define a greedy algorithm to select bitmap join indices that minimize processing cost and verify storage constraint. Finally, in order to evaluate the efficiency of our approach, we compare it with some existing techniques.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.6
• /
• pp.850-858
• /
• 2022

To deploy Gate Recurrent Units (GRU) on resource-constrained embedded devices, this paper presents a reconfigurable FPGA-based GRU accelerator that enables structured compression. Firstly, a dense GRU model is significantly reduced in size by hybrid quantization and structured top-k pruning. Secondly, the energy consumption on external memory access is greatly reduced by the proposed reuse computing pattern. Finally, the accelerator can handle a structured sparse model that benefits from the algorithm-hardware co-design workflows. Moreover, inference tasks can be flexibly performed using all functional dimensions, sequence length, and number of layers. Implemented on the Intel DE1-SoC FPGA, the proposed accelerator achieves 45.01 GOPs in a structured sparse GRU network without batching. Compared to the implementation of CPU and GPU, low-cost FPGA accelerator achieves 57 and 30x improvements in latency, 300 and 23.44x improvements in energy efficiency, respectively. Thus, the proposed accelerator is utilized as an early study of real-time embedded applications, demonstrating the potential for further development in the future.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.44 no.11
• /
• pp.82-88
• /
• 2007

In this paper, we present a pattern comparison algorithm for reducing fault candidate lists. The number of fault candidates determines the total fault simulation time. To decrease the total fault diagnosis time, the reduction of the number of fault candidates is essential. Critical path tracing determines fault candidate lists detected by a set of tests using a backtracing algorithm starting at the primary outputs of a circuit. The proposed algorithm reduces fault candidates comparing failing patterns with good patterns during critical path tracing process. As we reduce all fault candidates of the circuit to more accurately suspected fault candidates, we can greatly reduce fault simulation time. The proposed algorithm greatly increases simulation speed than that of a conventional backtracing method. The proposed algorithm is applicable to both combinational and sequential circuits. Experimental results on ISCAS#85 and ISCAS#89 benchmark circuits showed fault candidates are pruned and fault diagnosis time is also decreased in proportion to fault candidate decrease.

• Journal of Korea Spatial Information System Society
• /
• v.11 no.2
• /
• pp.151-161
• /
• 2009

In Location-Based Services (LBSs), users requesting a location-based query send their exact location to a database server and thus the location information of the users can be misused by adversaries. Therefore, a privacy protection method is required for using LBS in a safe way. In this paper, we propose a new cloaking region generation algorithm using grids for privacy protection in LBSs. The proposed algorithm creates a m inimum cloaking region by finding L buildings and then performs K-anonymity to search K users. For this, we make use of not only a grid-based index structure, but also an efficient pruning techniques. Finally, we show from a performance analysis that our cloaking region generation algorithm outperforms the existing algorithm in term of the size of cloaking region.

• Journal of Korea Spatial Information System Society
• /
• v.11 no.4
• /
• pp.39-46
• /
• 2009

Generally, Path-finding algorithms which use heuristic function may occur a problem of the increase of exploring cost in case of that there is no way determined by heuristic function or there are 2 way more which have almost same cost. In this paper, we propose an abstract graph for path-finding with dynamic information. The abstract graph is a simple graph as real road network is abstracted. The abstract graph is created by fixed-size cells and real road network. Path-finding with the abstract graph is composed of two step searching, path-finding on the abstract graph and on the real road network. We performed path-finding algorithm with the abstract graph against A* algorithm based on fixed-size cells on road network that consists of 106,254 edges. In result of evaluation of performance, cost of exploring in path-finding with the abstract graph is about 3~30% less than A* algorithm based on fixed-size cells. Quality of path in path-finding with the abstract graph is, However, about 1.5~6.6% more than A* algorithm based on fixed-size cells because edges eliminated are not candidates for path-finding.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.3
• /
• pp.1093-1104
• /
• 2014

The High Efficiency Video Coding (HEVC) is a new video coding standard that can provide much better compression efficiency than its predecessor H.264/AVC. However, it is computationally more intensive due to the use of flexible quadtree coding unit structure and more choices of prediction modes. In this paper, a fast intraframe coding scheme is proposed for HEVC. Firstly, a fast bottom-up pruning algorithm is designed to skip the mode decision process or reduce the candidate modes at larger block size coding unit. Then, a low complexity rough mode decision process is adopted to choose a small candidate set, followed by early DC and Planar mode decision and mode filtering to further reduce the number of candidate modes. The proposed method is evaluated by the HEVC reference software HM8.2. Averaging over 5 classes of HEVC test sequences, 41.39% encoding time saving is achieved with only 0.77% bitrate increase.

• Proceedings of the KSRS Conference
• /
• 2005.10a
• /
• pp.169-172
• /
• 2005

Modem SAR interferometry (IFSAR) sensors delivering intensity images and corresponding digital terrain model (DTM) allow for a thorough surface lineament interpretation with the all-weather day-night applicability. In this paper, an automatic linear-feature detection algorithm for high-resolution SAR images acquired in Taiwan is proposed. Methodologies to extract linear features consist of several stages. First, the image denoising techniques are used to remove the speckle noise on the raw image. In this stage, the Lee filter has been chosen because of its superior performance. After denoising, the Coefficient of Variation Detector is performed on the result images for edge enhancements and detection. Dilation and erosion techniques are used to reconnect the fragmented lines. The Hough transform, which is a special case of a more general transform known as Radon transform, is a suitable method for line detection in our analysis. Finally, linear features are extracted from the binary edge image. The last stage contains many substeps such as edge thinning and curve pruning.