• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.489-490
• /
• 2023

기계 학습에서 데이터 및 기능은 기계 학습의 상한을 결정한다.이러한 기능은 산업 생산에서 과도한 데이터 양과 유형으로 인해 상당한 추가 비용이 발생할 수 있다. 따라서 적절한 특징 처리 방법이 매우 중요해졌다. 대부분의 기존 특징 처리 방법은 특징 엔지니어링을 기능 검색 문제, 즉 모델 성능을 최적화할 수 있는 기능 변환 작업을 검색하는 것으로 추상화한다. 그러나 자동 특징 엔지니어링의 경우 검색량과 변환 조합의 수가 매우 많기 때문에 요인 분해 기반 모델을 사용하여 벡터 곱셈을 통해 상호 작용을 측정하면 조합 특징의 패턴을 자동으로 학습하는 방법이 특히 효율적이다. xDeepFM 은 명확한 방식으로 특징적인 상호작용을 생성하도록 설계된 새로운 Compressed Interaction Network (CIN)를 제안한다. 여기에 제시된 Low-rank Compressed Interaction Network(LRCIN )은 xDeepFM 접근 방식에서 CIN 네트워크의 단순화된 개선을 기반으로 하며 xDeepFM 에 주의 메커니즘을 추가하여 보다 정확하게 예측된다. 실험 결과에 따르면 모델은 계산 복잡성을 단순화할 뿐만 아니라 예측 정확도도 다른 모델보다 훨씬 우수한다.

• Journal of Information Processing Systems
• /
• v.17 no.2
• /
• pp.369-384
• /
• 2021

In a personalized product recommendation system, when the amount of log data is large or sparse, the accuracy of model recommendation will be greatly affected. To solve this problem, a personalized product recommendation method using deep factorization machine (DeepFM) to analyze user behavior is proposed. Firstly, the K-means clustering algorithm is used to cluster the original log data from the perspective of similarity to reduce the data dimension. Then, through the DeepFM parameter sharing strategy, the relationship between low- and high-order feature combinations is learned from log data, and the click rate prediction model is constructed. Finally, based on the predicted click-through rate, products are recommended to users in sequence and fed back. The area under the curve (AUC) and Logloss of the proposed method are 0.8834 and 0.0253, respectively, on the Criteo dataset, and 0.7836 and 0.0348 on the KDD2012 Cup dataset, respectively. Compared with other newer recommendation methods, the proposed method can achieve better recommendation effect.

• International journal of advanced smart convergence
• /
• v.12 no.2
• /
• pp.25-33
• /
• 2023

The purpose of this study is to evaluate the impact of intermediate features on FER performance. To achieve this objective, intermediate features were extracted from the input images at specific layers (FM1~FM4) of the pre-trained network (Resnet-18). These extracted intermediate features and original images were used as inputs to the vision transformer (ViT), and the FER performance was compared. As a result, when using a single image as input, using intermediate features extracted from FM2 yielded the best performance (training accuracy: 94.35%, testing accuracy: 75.51%). When using the original image as input, the training accuracy was 91.32% and the testing accuracy was 74.68%. However, when combining the original image with intermediate features as input, the best FER performance was achieved by combining the original image with FM2, FM3, and FM4 (training accuracy: 97.88%, testing accuracy: 79.21%). These results imply that incorporating intermediate features alongside the original image can lead to superior performance. The findings can be referenced and utilized when designing the preprocessing stages of a deep learning model in FER. By considering the effectiveness of using intermediate features, practitioners can make informed decisions to enhance the performance of FER systems.

• Ocean and Polar Research
• /
• v.33 no.spc3
• /
• pp.385-395
• /
• 2011

Time-series sediment traps were deployed at 1,000 m water depth of the northwestern subtropical Pacific from July 2009 to June 2010, with the aim of understanding temporal and spatial variations of sinking-particle fluxes. The opening and closing of the traps was synchronized at 18-day periods for 20 events. Total mass fluxes showed distinct seasonal variations with high values for the summer-fall seasons and relatively low values for winter-spring. This seasonal variation at two stations was characterized by a distinct difference in $CaCO_3$ fluxes between the two seasons. The enhanced $CaCO_3$ flux in the summer - fall seasons might be attributed to an increased planktonic foraminiferal flux. Total mass flux at FM10 station was nearly 50% higher than that at FM1 station. The difference in $CaCO_3$ fluxes between two stations contributed nearly 70% of the difference of total mass fluxes. The $CaCO_3$ flux was a major component controlling temporal and spatial variation of sinking - particle fluxes in the western subtropical Pacific Ocean.

• Korean Management Science Review
• /
• v.12 no.1
• /
• pp.19-34
• /
• 1995

The objective of this paper is to present a framework for an expert system for fault diagnosis in an FMS (Flexible Manufacturing Systyem). First, a system is analyzed structurally and functionally, giving the relationships between the system's components. These relationships, represented by strata, are are then stored in a deep knowledge base (DKB). Next, the specific knowledge, represented by echelons, about the symptoms and their probable causes for each component is stored in a shallow knowledge base (SKB) in the form of rule. When the fault diagnosis process begins, it starts to search the DKB and then the SKB, which is called hybrid reasoning in artificial intelligence.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.16 no.11
• /
• pp.7439-7446
• /
• 2015

The purpose of this study was to find the risk factors of injury by FMS and to investigate the effects of 12-weeks' combined training program on body imbalance, physical fitness, muscle strength and FMS score in woman rugby players of the national team. Fourteen subjects were woman rugby national players to participate in the 17th Incheon Asian Games. These players tested FMS and performed 12 weeks' (May~Aug, 2014) combined training program (4days, 120min${\geq}$). Statistical evaluation was undertaken using paired t-test (pre vs. post). The results of this study were as follows; Frist, the score of FMS test on Deep Squat, Hurdle Step, Active Straight Leg Raise, and Trunk Stability Push up were significantly increased after 12 weeks' combined training program (p<.05), and also Inline Lunge (p<.01) and Rotary Stability (p<.001) were significantly increased. However, Shoulder Mobility was not significantly increased(p=.104) although the tendency of increased was FMS score. Second, the sum of the entire item was significantly increased after combined training(p<.001). These results suggest that 12-weeks' combined training program has effect of improving FMS score and low-injury risk in woman rugby national players. Therefore, we consider that FMS have effect on prevention of athletic injury and improvement of athletic performance in woman rugby national players.

• The Korean Journal of Petroleum Geology
• /
• v.14 no.1
• /
• pp.1-11
• /
• 2008

As conventional oil and gas reservoirs become depleted, interests for oil sands has rapidly increased in the last decade. Oil sands are mixture of bitumen, water, and host sediments of sand and clay. Most oil sand is unconsolidated sand that is held together by bitumen. Bitumen has hydrocarbon in situ viscosity of >10,000 centipoises (cP) at reservoir condition and has API gravity between $8-14^{\circ}$. The largest oil sand deposits are in Alberta and Saskatchewan, Canada. The reverves are approximated at 1.7 trillion barrels of initial oil-in-place and 173 billion barrels of remaining established reserves. Alberta has a number of oil sands deposits which are grouped into three oil sand development areas - the Athabasca, Cold Lake, and Peace River, with the largest current bitumen production from Athabasca. Principal oil sands deposits consist of the McMurray Fm and Wabiskaw Mbr in Athabasca area, the Gething and Bluesky formations in Peace River area, and relatively thin multi-reservoir deposits of McMurray, Clearwater, and Grand Rapid formations in Cold Lake area. The reservoir sediments were deposited in the foreland basin (Western Canada Sedimentary Basin) formed by collision between the Pacific and North America plates and the subsequent thrusting movements in the Mesozoic. The deposits are underlain by basement rocks of Paleozoic carbonates with highly variable topography. The oil sands deposits were formed during the Early Cretaceous transgression which occurred along the Cretaceous Interior Seaway in North America. The oil-sands-hosting McMurray and Wabiskaw deposits in the Athabasca area consist of the lower fluvial and the upper estuarine-offshore sediments, reflecting the broad and overall transgression. The deposits are characterized by facies heterogeneity of channelized reservoir sands and non-reservoir muds. Main reservoir bodies of the McMurray Formation are fluvial and estuarine channel-point bar complexes which are interbedded with fine-grained deposits formed in floodplain, tidal flat, and estuarine bay. The Wabiskaw deposits (basal member of the Clearwater Formation) commonly comprise sheet-shaped offshore muds and sands, but occasionally show deep-incision into the McMurray deposits, forming channelized reservoir sand bodies of oil sands. In Canada, bitumen of oil sands deposits is produced by surface mining or in-situ thermal recovery processes. Bitumen sands recovered by surface mining are changed into synthetic crude oil through extraction and upgrading processes. On the other hand, bitumen produced by in-situ thermal recovery is transported to refinery only through bitumen blending process. The in-situ thermal recovery technology is represented by Steam-Assisted Gravity Drainage and Cyclic Steam Stimulation. These technologies are based on steam injection into bitumen sand reservoirs for increase in reservoir in-situ temperature and in bitumen mobility. In oil sands reservoirs, efficiency for steam propagation is controlled mainly by reservoir geology. Accordingly, understanding of geological factors and characteristics of oil sands reservoir deposits is prerequisite for well-designed development planning and effective bitumen production. As significant geological factors and characteristics in oil sands reservoir deposits, this study suggests (1) pay of bitumen sands and connectivity, (2) bitumen content and saturation, (3) geologic structure, (4) distribution of mud baffles and plugs, (5) thickness and lateral continuity of mud interbeds, (6) distribution of water-saturated sands, (7) distribution of gas-saturated sands, (8) direction of lateral accretion of point bar, (9) distribution of diagenetic layers and nodules, and (10) texture and fabric change within reservoir sand body.

• Journal of Microbiology and Biotechnology
• /
• v.19 no.10
• /
• pp.1206-1212
• /
• 2009

Growth rates, photosystem II photosynthesis, and the levels of chlorophyll $\alpha$ and secondary metabolites of Chlorella ovalis were estimated to determine if they were enhanced by the addition of swine urine (BM) or cow compost water (EP) that had been fermented by soil bacteria to deep seawater (DSW) in an attempt to develop media that enabled batch mass culture at lower costs. Growth of C. ovalis in f/2, f/2-EDTA+BM60%, DSW+BM30%, and DSW+EP60% was enhanced and maintained in the log phase of growth for 16 days. The cell densities of C. ovalis in DSW+EP60% ($4.1{\times}10^6$ Cells/ml) were higher than those of f/2 ($2.9{\times}10^6$ Cells/ml), f/2-E+BM60% ($3.7{\times}10^6$ Cells/ml), and DSW+BM30% ($2.7{\times}10^6$ Cells/ml). The growth rate was also more favorable for C. ovalis cultured in DSW+EP60% ($0.15\;day^{-1}$) than that of C. ovalis cultured in the control medium (f/2) ($0.12\;day^{-1}$). Furthermore, the chlorophyll a concentration of C. ovalis cultured in DSW+EP60% (4.56 mg/l) was more than 2-fold greater than that of C. ovalis cultured in f/2 (2.35 mg/l). Moreover, the maximal quantum yields of photo system II at 470 nm (Fv/Fm) were significantly higher in organisms cultured at f/2-E+BM60% (0.53) and DSW+EP60% (0.52) than in the other treatment groups. Finally, Fourier transformation infrared (FT-IR) spectroscopy revealed that C. ovalis grown in DSW+EP60% had more typical peaks and various biochemical pool shifts than those grown in other types of media. Taken together, the results of this study indicate that the use of DSW+EP60% to culture C. ovalis can reduce maintenance expenses and promote higher yields.

• Journal of the Korean Academy of Clinical Electrophysiology
• /
• v.4 no.1
• /
• pp.63-83
• /
• 2006

This study conducted the following experiment to examine anesthetic effects of 500 kHz ultrasound with lidocaine. Clinical experiment was conducted to compare local anesthetic effects by ultrasound frequencies and drugs with 40 normal adults and it divided subjects into experimental group (I) applying only ultrasound, ointment group (II) applying only lidocaine, phonophoresis group of 1 MHz (III), phonophoresis group of 500 kHz lidocaine (IV) for pain threshold and nerve conduct experiments. Mechanical threshold was measured with von Frey monofilament, thermal pain threshold with halogen lamp and digital thermometer, action potential in median nerve with diagnostic electromyography before and after treatment, and at 30 min., 60 min., 90 min., and 120 min. after treatment. Results of this study can be summarized as follows. Topical anesthesia experiment showed that pain threshold of phonophoresis groups was significantly increased, but there were little differences in ultrasound frequency and drugs among phonophoresis groups. Conduction anesthesia experiment showed that phonophoresis group of 500 kHz using lidocaine had significant difference in pain threshold and sensory nerve conduction compared to ointment group and ultrasound group, but there was no great difference from other phonophresis groups and light nerve block effect was found. It was considered that application of 500 kHz ultrasound using lidocaine will be more effective in deep anesthesia or nerve block than 1 MHz ultrasound. However, researches considering various frequencies, intensities and application hours in low frequency areas including kHz ultrasound are needed to increase deep permeation of drugs.

• Journal of Bio-Environment Control
• /
• v.17 no.4
• /
• pp.252-260
• /
• 2008

This study was conducted to observe the characteristics of mineral nutrient uptake of single-node cutting rose 'Versilla' and to determine optimal nutrient solution control method for soilless culture of 'Versilla' in a closed hydroponic system. Nutrient solution was managed by five different control methods: macro- and micro-element control in aeroponic system (M&M), macroelement control in aeroponic system (M), nutrient solution supplement in aeroponic system (S); electrical conductivity (EC) control in aeroponic system(EC-A); EC control in deep flow technique system(EC-D). The concentration of $NO_3$-N exceeds optimal range whereas P and Mg decreased at the later stage of plant growth with the EC control method, EC-A and EC-D. The overall mineral nutrient content increased with S. On the other hand the nutrient content at the root environment was maintained optimal with M&M and M. The nutrient solution control methods had significantly effect on the cut-flower quality. In the M&M and M, flower length, fresh weight and root activity were higher than those with the other mineral nutrients control methods. The maximal efficiency of photochemistry (Fv/Fm) was higher for M&M, M and S than that with EC-A and EC-D. Based on the above results, it is highly recommended to control nutrient solution by mineral nutrient control methods (M&M and M) in a closed hydroponic system for single-node cutting rose, 'Versillia'.