• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Journal of Korea Water Resources Association
• /
• v.42 no.12
• /
• pp.1079-1089
• /
• 2009

The objective of this study is to develop the accurate, robust and high resolution two-dimensional numerical model that solves the computationally difficult hydraulic problems, including the wave front propagation over dry bed and abrupt change in bathymetry. The developed model in this study solves the conservative form of the two-dimensional shallow water equations using an unsplit finite volume scheme and HLLC approximate Riemann solvers to compute the interface fluxes. Bed-slope term is discretized by the divergence theorem in the framework of FVM for application of unsplit scheme. Accurate and stable SGM, in conjunction with the MUSCL which is second-order-accurate both in space and time, is adopted to balance with fluxes and source terms. The exact C-property is shown to be satisfied for balancing the fluxes and the source terms. Since the spurious oscillations in second-order schemes are inherent, an efficient slope limiting technique is used to supply TVD property. The accuracy, conservation property and application of developed model are verified by comparing numerical solution with analytical solution and experimental data through the simulations of one-dimensional dam break flow without bed slope, steady transcritical flow over a hump and two-dimensional dam break flow with a constriction.

• The Transactions of the Korea Information Processing Society
• /
• v.5 no.4
• /
• pp.1033-1043
• /
• 1998

In this paper, we present an improved multicasting path assignment algorithm based on the minimum cost spanning tree. In the method presented in this paper, a multicasting path is assigned preferentially when a node to be received is found among the next degree nodes of the searching node in the multicasting path assignment of the constrained steiner tree (CST). If nodes of the legacy group exist between nodes of the new group, a new path among the nodes of new group is assigned as long as the nodes may be excluded from the new multicasting path assignment taking into consideration characteristics of nodes in the legacy group. In assigning the multicasting path additionally, where the source and destination nodes which can be set for the new multicasting path exist in the domain of identical network (local area network) and conditions for degree constraint are satisfied, a method of producing and assigning a new multicasting path is used. The results of comparison of CST with MCSTOP, MCSTOp algorithm enhanced performance capabilities about the communication cost, the propagation delay, and the computation time for the multicasting assignment paths more than CST algorithm. Further to this, research activities need study for the application of the international standard protocol(multicasting path assignment technology in the multipoint communication service (MCS) of the ITU-T T.120).

• Journal of KIISE:Software and Applications
• /
• v.33 no.12
• /
• pp.1022-1030
• /
• 2006

The location awareness is an important part of many ubiquitous computing systems, but a perfect location system does not exist yet in spite of many researches. Among various location tracking systems, we choose the RFID system due to its wide applications. However, the sensed RSSI signal is too sensitive to the direction of a RFID reader antenna, the orientation of a RFID tag, the human interference, and the propagation media situation. So, the existing location tracking method in spite of using the particle filter is not working well. To overcome this shortcoming, we suggest a robust location tracking method with a double layered structure, where the first layer coarsely estimates a tag's location in the block level using a regression technique or the SVM classifier and the second layer precisely computes the tag's location, velocity and direction using the particle filter technique. Its layered structure improves the location tracking performance by restricting the moving degree of hidden variables. Many extensive experiments show that the proposed location tracking method is so precise and robust to be a good choice for implementing the location estimation of a person or an object in the ubiquitous computing. We also validate the usefulness of the proposed location tracking method by implementing it for a real-time people monitoring system in a noisy and complicate workplace.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.53 no.7
• /
• pp.17-26
• /
• 2016

This work proposes a 12b 60MS/s 0.18um CMOS Flash-SAR ADC for various systems such as wireless communications and portable video processing systems. The proposed Flash-SAR ADC alleviates the weakness of a conventional SAR ADC that the operation speed proportionally increases with a resolution by deciding upper 4bits first with a high-speed flash ADC before deciding lower 9bits with a low-power SAR ADC. The proposed ADC removes a sampling-time mismatch by using the C-R DAC in the SAR ADC as the combined sampling network instead of a T/H circuit which restricts a high speed operation. An interpolation technique implemented in the flash ADC halves the required number of pre-amplifiers, while a switched-bias power reduction scheme minimizes the power consumption of the flash ADC during the SAR operation. The TSPC based D-flip flop in the SAR logic for high-speed operation reduces the propagation delay by 55% and the required number of transistors by half compared to the conventional static D-flip flop. The prototype ADC in a 0.18um CMOS demonstrates a measured DNL and INL within 1.33LSB and 1.90LSB, with a maximum SNDR and SFDR of 58.27dB and 69.29dB at 60MS/s, respectively. The ADC occupies an active die area of $0.54mm^2$ and consumes 5.4mW at a 1.8V supply.

• KSBB Journal
• /
• v.13 no.3
• /
• pp.231-237
• /
• 1998

An attenuated Japanese encephalitis virus (JEV) clone SA-14-14-2(Vero) was obtained through successive adaptation of a primary cell adapted strain, SA-14-14-2(PDK) to Vero cell, a continuous line of monkey kidney cells. The virus titer reached above the 107 plaque forming unit (pfu) per mL of culture supernatant with 3 passages in Vero cells and was maintained close to this level in the further passages. The optimum temperature for the virus growth was $35^{\circ}C$. Growth pattern of the virus indicated that optimum time for the virus harvest is 4 days post infection and the virus accomplished rapid initial propagation even in medium containing no serum supplement. The roller bottle (RB) system and the spinner flask (SF) system using micro-carrier (Cytodex-1) for the JEV cultivation were explored. When RB, SF, and T-flask culture system were compared, there was no significant difference in virus yield. Furthermore, the results indicated that virus could be harvested multiple times from 3 days to 9 days post infection; neither severe cytopathic effect (CPE) in the infected cells nor the decrease in the titer was observed on duration of 9 days.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.43 no.5 s.311
• /
• pp.36-43
• /
• 2006

In this paper we present a new high-speed parallel multiplier for Performing the bit-parallel multiplication of two polynomials in the finite fields $GF(2^m)$. Prior to construct the multiplier circuits, we consist of the MOD operation part to generate the result of bit-parallel multiplication with one coefficient of a multiplicative polynomial after performing the parallel multiplication of a multiplicand polynomial with a irreducible polynomial. The basic cells of MOD operation part have two AND gates and two XOR gates. Using these MOD operation parts, we can obtain the multiplication results performing the bit-parallel multiplication of two polynomials. Extending this process, we show the design of the generalized circuits for degree m and a simple example of constructing the multiplier circuit over finite fields $GF(2^4)$. Also, the presented multiplier is simulated by PSpice. The multiplier presented in this paper use the MOD operation parts with the basic cells repeatedly, and is easy to extend the multiplication of two polynomials in the finite fields with very large degree m, and is suitable to VLSI. Also, since this circuit has a low propagation delay time generated by the gates during operating process because of not use the memory elements in the inside of multiplier circuit, this multiplier circuit realizes a high-speed operation.

• Journal of the Korea Concrete Institute
• /
• v.23 no.2
• /
• pp.219-224
• /
• 2011

In this study, a piezoelectric smart sensor that can be embedded inside of concrete structures is developed to investigate the early stage of concrete curing. A waterproof coating is used to protect the piezoelectric sensor from moistures of concrete mixture. Also, a mortar case is utilized to encapsulate the sensor to protect it from impact loads. To estimate the strength of concrete, a self-sense guided-wave actuated sensing technique is applied. In the guided wave, its velocity is varied according to the mechanical properties of concrete such as modulus of elasticity. Because modulus of elasticity directly affects the strength of concrete, the guidedwave's velocity also affects the concrete strength development. To verify the feasibility of using the proposed approach, the smart sensor was embedded into a 100MPa concrete cylinder and the self-sense guided wave is continuously measured throughout the curing process. The measurements showed that the propagation time (TOF) of the measured guided waves gradually decreased as the curing age increased. Especially, at the early age of the curing process, the variation of the TOF was very significant. Furthermore, the results showed that there is a linear relationship between the TOF of the self-sense guided waves and the strength of concrete existed. It is safe to conclude that the proposed approach can be used very effectively in monitoring of the strength development of high strength concrete structures.

• The Journal of Korean Academy of Prosthodontics
• /
• v.38 no.5
• /
• pp.595-605
• /
• 2000

Dental ceramics have good aesthetics, biocompatibility, low thermal conductivity, abrasion resistance, and color stability. However poor resistance to fracture and shrinkage during firing process have been limiting factors in their use, particularly in multiunit ceramic restorations. A new method for making all-ceramic crowns that have high strength and low processing shrinkage has been developed and is referred to as the Vita In-Ceram method. This study was performed to investigate the effect of $CeO_2$ addition in borosilicate glasses on the strength of alumina-glass composites. Porous alumina compacts were prepared by slip casting and sintered at $1,100^{\circ}C$ for 2 hours. Dense composites were made by infiltration of molten glass into partially sintered alumina at $1,140^{\circ}C$ for 4 hours. Specimens were polished sequentially from #800 to #2000 diamond disk. and the final surface finishing on the tensile side was received an additional polishing sequence through $1{\mu}m$ diamond paste. Biaxial flexure test was conducted by using ball-on-three-ball method at a crosshead speed of 0.5mm/min. To examine the microstructural aspect of crack propagation in the alumina-glass composites, Vickers-produced indentation crack was made on the tensile surface at a load of 98.0 N and dwell time of 15 sec, and the radial crack patterns were examined by an optical microscope and a scanning electron microscope. The results obtained were summarized as follows; 1. The porosity rates of partially sintered alumina decreased with the rising of firing temperature. 2. The maximum biaxial flexure strength of 423.5MPa in alumina-glass composites was obtained with an addition of 3 mol% $CeO_2$ in glass composition and strength values showed the aspect of decrease with the increase of $CeO_2$ content. 3 The biaxial flexure strength values of alumina-glass composites were decreased with rising the firing temperature. 4. Observation of the fracture surfaces of alumina-glass composites indicated that the enhancement of strength in alumina-glass composites was due to the frictional or geometrical inter-locking of rough fracture surfaces and ligamentary bridging by intact islands of materials left behind the fracture front.

• Journal of the Korean Society for Nondestructive Testing
• /
• v.20 no.4
• /
• pp.276-284
• /
• 2000

The acoustic emission (AE) behavior of reinforced concrete beams tested under flexural loading was investigated to characterize and identify the source of damage. This research was aimed at identifying the characteristic AE response associated with micro-crack development, localized crack propagation, corrosion, and debonding of the reinforcing steel. Concrete beams were prepared to isolate the damage mechanisms by using plain, notched-plain, reinforced, and corroded-reinforced specimens. The beams were tested using four-point cyclic step-loading. The AE response was analyzed to obtain key parameters such as the time history of AE events, the total number and rate of AE events, and the characteristic features of the waveform. Initial analysis of the AE signal has shown that a clear difference in the AE response is observed depending on the source of the damage. The Felicity ratio exhibited a correlation with the overall damage level, while the number of AE events during unloading can be an effective criterion to estimate the level of corrosion distress in reinforced concrete structures. Consequently, AE measurement characterization appears to provide a promising approach for estimating the level of deterioration in reinforced concrete structure.