• ETRI Journal
• /
• v.33 no.4
• /
• pp.652-655
• /
• 2011

In this letter, we propose a new compression method for a high dimensional support vector machine (SVM). We used singular value decomposition (SVD) to compress the norm part of a radial basis function SVM. By deleting the least significant vectors that are extracted from the decomposition, we can compress each vector with minimized energy loss. We select the compressed vector dimension according to the predefined threshold which can limit the energy loss to design criteria. We verified the proposed vector compressed SVM (VCSVM) for conventional datasets. Experimental results show that VCSVM can reduce computational complexity and memory by more than 40% without reduction in accuracy when classifying a 20,958 dimension dataset.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.540-542
• /
• 2023

Due to the serious issues posed by facial manipulation technologies, many researchers are becoming increasingly interested in the identification of face forgeries. The majority of existing face forgery detection methods leverage powerful data adaptation ability of neural network to derive distinguishing traits. These deep learning-based detection methods frequently treat the detection of fake faces as a binary classification problem and employ softmax loss to track CNN network training. However, acquired traits observed by softmax loss are insufficient for discriminating. To get over these limitations, in this study, we introduce a novel discriminative feature learning based on Vision Transformer architecture. Additionally, a separation-center loss is created to simply compress intra-class variation of original faces while enhancing inter-class differences in the embedding space.

• Journal of Korea Multimedia Society
• /
• v.19 no.2
• /
• pp.360-374
• /
• 2016

Since WSNs (Wireless Sensor Networks) applied to their application areas such as smart home, smart factory, environment monitoring, etc., depend on sensor data, the sensor data is the most important among WSN components. The resources of each node consisting of WSN are extremely limited in energy, hardware and so on. Due to these limitation, communication failure probabilities become much higher and the communication failure causes data loss to occur. For this reason, this paper proposes 2MC (Maximum/Minimum Compression) that is a method to compress sensor data by selecting circular queue-based maximum/minimum sensor data values. Our proposed method reduces sensor data losses and value errors when they are recovered. Experimental results of 2MC method show the maximum/minimum 35% reduction efficiency in average sensor data accumulation error rate after the 3 times compression, comparing with CQP (Circular Queue Compression based on Period) after the compressed data recovering.

• Transactions of the Korean Society of Automotive Engineers
• /
• v.16 no.2
• /
• pp.35-42
• /
• 2008

The compressive residual stress due to shot peening process can increase the intrinsic fatigue strength of surface and therefore would be beneficial in reducing the probability of fatigue damage. However, it was not known that the effect of shot peening on corrosion environment. In this study, the influence of shot peening and corrosion condition on corrosion property was investigated on immersed in 3.5% NaCl, 10% $HNO_3$+3% HF, 6% $FeCl_3$. The immersion test was performed with two kind of specimens. The immersion test periods were carried out on performed 360days. Corrosion potential and weight loss were investigated from experimental results. From these results, the effect of shot peening on the corrosion characteristics was evaluated.

• Proceedings of the KSME Conference
• /
• 2008.11b
• /
• pp.2429-2433
• /
• 2008

A rotary vane blower was developed as an air supply system for fuel cell application. As one way of improving the blower efficiency, a roller was adapted between vanes and cylinder housing. The performance of blower was investigated experimentally. The blower power input was about 115W to compress the air at normal atmospheric condition to 0.2 bar with the flow rate of 140 NLPM, resulting in the blower overall efficiency of 43%. After 400 hours of operation, the performance of blower was not changed. The result showed that developed blower was confirmed to be suitable for fuel cell application.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.06a
• /
• pp.84-85
• /
• 2019

MPEG-NNR (Compression of Neural Network for Multimedia Content Description and Analysis) aims to define a compressed and interoperable representation of trained neural networks. In this paper, we present a low-rank approximation to compress a CNN used for image compression, which is one of MPEG-NNR use cases. In the presented method, the low-rank approximation decomposes one 2D kernel matrix of weights into two 1D kernel matrix values in each convolution layer to reduce the data amount of weights. The evaluation results show that the model size of the original CNN is reduced to half as well as the inference runtime is reduced up to about 30% with negligible loss in PSNR.

• Journal of the Semiconductor & Display Technology
• /
• v.21 no.4
• /
• pp.156-162
• /
• 2022

Image compression plays an important role in encoding and improving various forms of images in the digital era. Recent researches have focused on the principle of deep learning as one of the most exciting machine learning methods to show that it is good scheme to analyze, classify and compress images. Various neural networks are able to adapt for image compressions, such as deep neural networks, artificial neural networks, recurrent neural networks and convolution neural networks. In this review paper, we discussed how to apply the rule of deep learning to obtain better image compression with high accuracy, low loss-ness and high visibility of the image. For those results in performance, deep learning methods are required on justified manner with distinct analysis.

• The Korean Journal of Pain
• /
• v.11 no.2
• /
• pp.346-349
• /
• 1998

The technique of the stellate ganglion block is widely used as it is relatively simple and safe. But it can cause severe complications because there are major blood vessels and nerves around the stellate ganglion. We practiced CPR because of the respiratory failure caused by severe hematoma in the neck following the stellate ganglion block. A 46-year-old male patient admitted to ENT department because of the both sudden sensorineural hearing loss that happened after URI. He was referred to Pain Clinic for further evaluation and treatment. We decided to block the stellate ganglion. We injected 6ml of 0.5% mepivacaine on both sides of the stellate ganglion. There were no blood aspiration and abnormal vital signs during the 30 minute observation, either. Three hours after he went to the private room, he had pain and edema in his neck, but no respiratory defficulty. But later, respiratory failure was suddenly followed. So we practiced CPR. We confirmed severe hematomas in the neck through CT scanning. Hematomas is removed and the ruptured blood vessels which is supposed to be muscular branch of vertebral artery is ligated under general anesthesia. The patient was discharged from hospital after the treatment of pneumonia and duodenal ulcer as complications. We recommand you to compress the block site more than five minutes and not to prick with the needle several times at one point to prevent the formation of hematomas.

• Proceedings of the KOSOMBE Conference
• /
• v.1994 no.12
• /
• pp.31-38
• /
• 1994

This paper proposed signal decomposition and multiresolution representation through wavelet transform using wavelet orthonormal basis. And it suggested most appropriate filter for scaling function in multiresoltion representation and compared two compression method, arithmetic coding and Huffman coding. Results are as follows 1. Daub18 coefficient is most appropriate in computing time, energy compaction, image quality. 2. In case of image browsing that should be small in size and good for recognition, it is reasonable to decompose to 3 scale using pyramidal algorithm. 3. For the case of progressive transmittion where requires most grateful image reconstruction from least number of sampls or reconstruction at any target rate, I embedded the data in order of significance after scaling to 5 step. 4. Medical images such as information loss is fatal have to be compressed by lossless method. As a result from compressing 5 scaled data through arithmetic coding and Huffman coding, I obtained that arithmetic coding is better than huffman coding in processing time and compression ratio. And in case of arithmetic coding I could compress to 38% to original image data.

• Journal of Advanced Marine Engineering and Technology
• /
• v.27 no.7
• /
• pp.889-898
• /
• 2003

The soot yield is studied by a premixed propane-oxygen-inert gas combustion in a specially designed disk-type constant-volume combustion chamber to investigate the effect of turbulence on soot formation. Premixtures are simultaneously ignited by eight spark plugs located on the circumference of chamber at 45 degree intervals in order to observe the soot formation under high pressures and high temperatures. The eight flames converged compress the end gases to a high pressure. The laser schlieren and direct flame photographs for observation field with 10 mm in diameter are taken to examine into the behaviors of flame front and gas flow in laminar and turbulent combustion. The soot volume fraction in the chamber center during the final stage of combustion at the highest pressure is measured by the in situ laser extinction technique and simultaneously the corresponding burnt gas temperature by the two-color pyrometry method. It is found that the soot yield of turbulent combustion decreases in comparison with that of laminar combustion because the burnt gas temperature increases with the drop of heat loss.