Earticle

연합학습(Federated Learning)은 로컬데이터 기반의 학습 파라미터를 서버로 보내어 파라미터를 업데이트하는 방식의 분산 학습 방식으로써 로컬 데이터 유출을 방지하면서 모델을 공동 학습할 수 있는 구조를 제공한다. 연합학 습은 로컬 데이터의 유출은 방지하지만 파라미터 전송으로 인한 네트워크 부하와 파라미터에 대한 역복원 공격 (Inversion Attack)에 취약하다는 한계가 있다. 본 연구는 연합학습에서 파라미터 전송 효율성과 보안성을 향상시 키기 위해 확장 양자화 방식을 적용한 성능 분석을 다룬다. 기존의 gradient 클리핑, 차등 프라이버시, 양자화 기법 은 역복원 공격에 대한 보안성을 높이지만 전송량 절감 효과에는 제한적이다. 본 연구는 양자화 시 값의 범위를 확 장하여 양자화를 수행함으로써 유효 파라미터값을 줄이는 확장 양자화 기법을 적용하여 보았고, 다양한 확장 비율과 양자화 레벨에 따른 전송량 절감 및 역복원 공격 방어 성능을 실험적으로 분석했으며, 실험 결과, 확장 양자화는 약 간의 정확도 손실을 감수하는 대신 전송량을 효과적으로 줄이고 보안성을 향상시킬 수 있음을 확인하였다.

Federated Learning is distributed learning method in which locally trained parameters are sent to a server for parameter updates, enabling collaborative model training while preventing the leakage of local data. Although it protects local data from exposure, Federated Learning still suffers from limitations such as network overhead caused by parameter transmission and vulnerability to inversion attacks on the transmitted parameters. This study analyzes the performance of applying an extended quantization method to improve both parameter transmission efficiency and security in Federated Learning. While existing techniques such as gradient clipping, differential privacy, and standard quantization can enhance security against inversion attacks, their effectiveness in reducing transmission volume is limited. In this work, we applied an extended quantization technique that reduces the number of effective parameter values by expanding the quantization range, and we experimentally analyzed its impact on transmission reduction and defense against inversion attacks across various expansion ratios and quantization levels. Experiments show that extended quantization can effectively reduce transmission volume and improve security, with the cost of small loss in model accuracy.

Owing to the continuous industrialization since 1970s, our country has coped with serious urbanization issues, particularly in the shortage of houses and lands, and has attempted to effectively solve this problem by developing high-rise and high-density multistory residences. Recently, the concern about developing residence on the mountainous areas and the sloping sites to meet the rising demand of new housing has been gradually growing. Once a slope developed to a residential site, it has various advantages over building a level site such as excellent view and sunlight, and diverse space layout. Hence, it will provide comfortable residential circumstances if more residences are developed on the sloping sites with their uniqueness and this will possibly solve those serious issues associated with scarcity of housing in our country's downtown area. In this context, this research analyzed the design characteristics applied in high-rise apartment housing built on the slopes, especially in relation to the gradient of slope.

High-risk pregnancy endangers both mother and baby, with one maternal death every two minutes in 2023. This study has proposed three Gradient Boosting models?GB-Base, GB-SMOTE, and ZOA-GB?using the West Lombok Pregnancy Risk Prediction Dataset. GB-Base and GB-SMOTE have achieved 90.46% and 90.28% accuracy, while ZOA-GB, using 10 selected features, has reached 88.89%. GB-SMOTE has shown the best performance with an F-score of 84.41%. SHAP has identified Maternal Age, Hemoglobin, and Parity as key features, and DiCE has validated feature-driven prediction control. The study is limited by a single-source dataset, the absence of external-validation, and unexplored optimizers.

Objective To investigate the relationship between the shoulder gradient and acromiohumeral interval of both shoulders in patients with unilateral shoulder impingement syndrome. Method Using the angulometer, we measured the shoulder gradient in patients with unilateral shoulder impingement syndrome in a standing position. Using the radiography, we measured the acromiohumeral interval and the angle between a vertical line and a line connecting a superior angle with an inferior angle of the scapula. Results In patients with unilateral shoulder impingement syndrome, the frequency of shoulder impingement syndrome was 76.2% (16 of 21) on the side of the relatively lower shoulder. The mean acromiohumeral interval on the side of the lower shoulder was 10.03±1.28 mm, compared with 10.46±1.50 mm for the higher shoulder. Th e angle between a vertical line and a line connecting a superior angle with an inferior angle of the scapular of the side of the lower shoulder was -0.31±3.73 degrees, compared with 3.85±4.42 degrees for the higher shoulder. Conclusion The frequency of shoulder impingement syndrome was significantly higher on the side of the relatively lower shoulder, and there is no signifi cant diff erence in the acromiohumeral interval between the side of the lower shoulder and that of the higher shoulder. In patients with unilateral shoulder impingement syndrome,the scapular on the side of lower shoulder was more rotated downward than on the side of the higher shoulder.

DDPG(Deep Deterministic Policy Gradient)알고리즘은 인공신경망과 강화학습을 사용하여 학습하는 알고리즘이다. 최근많은 연구가 이루어지고 있는 강화학습과 관련된 연구 중에서도 DDPG 알고리즘은 오프폴리시로 학습하기 때문에 잘못된 행동이 누적되어 학습에 영향을 미치는 경우를 방지하는 장점이 있다. 본 연구에서는 DDPG 알고리즘을 응용하여 자전거를 자율주행 하도록 제어하는 실험을 진행하였다. 다양한 환경을 설정하여 시뮬레이션을 진행하였고 실험을 통해서 사용된 방법이시뮬레이션 상에서 안정적으로 동작함을 보였다.

The Deep Deterministic Policy Gradient (DDPG) algorithm is an algorithm that learns by using artificial neural network s and reinforcement learning. Among the studies related to reinforcement learning, which has been recently studied, the D DPG algorithm has an advantage of preventing the cases where the wrong actions are accumulated and affecting the learn ing because it is learned by the off-policy. In this study, we experimented to control the bicycle autonomously by applyin g the DDPG algorithm. Simulation was carried out by setting various environments and it was shown that the method us ed in the experiment works stably on the simulation.

NGBoost has shown promising results in probabilistic and point estimation tasks. However, it is vague still whether this method can be scalable to neural architecture system since its base learner is based on decision trees. To resolve this, we design a Neural-NGBoost framework by replacing the base learner with lightweight neural networks and introducing joint gradient estimation for boosting procedure. Based on natural gradient boosting, we iteratively update the neural based learner by inferring natural gradient and update the parameter score with its probabilistic distribution. Experimental results show Neural-NGBoost achieves superior performance across various datasets compared to other boosting methods.

We have evaluated and compared of gradient echo and spin echo EPI for compensating about deeply distortion aspect in case of post-operation patients in magnetic resonance image. A total of 100 patients were performed on 3.0 T(GE Signa Excite E2, USA) with 8ch head coil. As a result of analysis, The SNRs of whiter and gray matter areas were 36.74 ± 06 and 39.96 ± 09 in the gradient echo EPI, the SNRs which white and gray matter areas were slightly higher than gradient echo EPI(P<0.005, paired student t-test). It was 46.24 ± 11 and 51.38 ± 13 in gradient and spin echo EPI, respectively. The signal intensity in the whiter and gray matter areas also were 87.33 ± 15.24 and 140.66 土 13.45 in the gradient echo EPI techniques, The signal intensity of gradient echo EPI showed higher values compared to spin echo EPI. Otherwise, gradient echo EPI technique is distortion enough to operation wound and edge of the image, while spin echo EPI technique did not appear almost. In this point, the spin echo EPI technique, after surgical operation according to patient state beside gradient echo EPI techniques that signalbeside gradient echo EPI techniques that signal intensity is high and patient's case which image distortion is serious by metal etc, will be provide the useful information in adults and pediatric patients.

To evaluate the accuracy and extent in the localization of cerebral motor cortex activation using a gradient- echo echo planar imaging(GE-EPI) compared to spin-echo echo planar imaging(SE-EPI) on 3T MR imaging. Functional MR imaging of cerebral motor cortex activation was examined in GE-EPI and SE-EPI in five healthy male volunteers. A right finger movement was accomplished with a paradigm of 6 task and rest periods and the cross-correlation was used for a statistical mapping algorithm. We evaluated any sorts of differences of the time series and the signal intensity changes between the rest and task periods obtained with two techniques. The qualitative analysis was distributed with activation sites of large veins and small veins by using two techniques and was found that both the techniques were clinically useful for delineating large veins and small veins in fMRI. Signal intensity change of the rest and activation periods provided similar activations in both methods(GE-EPI : 0.93±0.11, SE-EPI : 0.80±0.15) but 仕le signal intensity in GE-EPI(133.95±15.76) was larger than in SE-EPI(74.5± 18.90). The average SNRs of EPI raw data were higher at SMA in SE-EPI(48.54±12.37) than GE-EPI(41.4±12.54) and at Ml in SE-EPI(43.24±11.77) than GE_EPI(38.27±6.53). The localization of activation voxds of the GE-EPI showed a larger vein but the SE-EPI generally showed small vein. Then the analysis results of the two techniques were used for a statistical paired student t-test. SE-EPI was found clinically useful for localizing the cerebral motor cortex activation on 3.0T, but showed a little different activation patterns compared to a GE-EPI. In conclusion, SE-EPI may be feasible and can detect true cortical activation from capillaries and GE-EPI can obtain the large veins in the motor cortex activation on 3T MR imaging.

Regression analysis is one of the most widely utilized methods because of its adaptability and simplicity. Recently, the machine learning (ML) approach, which is one aspect of regression methods, has been gaining attention from researchers, including social science, but there are only a few studies that compared the traditional approaches with the ML approach. This study was conducted to explore the usefulness of the ML approach by comparing the ordinary least square estimate (OLS), the stochastic gradient descent algorithm (SGD), and the support vector regression (SVR) with a model predicting and explaining the tuberculosis screening intention. The optimized models were evaluated by four aspects: computational speed, effect and importance of individual predictor, and model performance. The result demonstrated that each model yielded a similar direction of effect and importance in each predictor, and the SVR with the radial kernel had the finest model performance compared to its computational speed. Finally, this study discussed the usefulness and attentive points of the ML approach when a researcher utilizes it in the field of communication.

This study aims to investigate whether the gradient effect of OCP (the Obligatory Contour Principle) can mirror the co-occurrence restrictions across a syllable boundary in English phonotactics. For this purpose, this study collected the data from the Carnegie Mellon University Pronouncing Dictionary and employed the Observed over Expected (O/E) analysis which is introduced by Pierrehumbert (1993) to measure the OCP effects. As a result, our finding is that the co-occurrence of the two consonants sharing the same manner feature across a syllable boundary is strongly restricted by OCP. In terms of OCP-place, the effect plays a role in an opposite way in that the co-occurrence of two consonants sharing the same place features are more likely to be represented than other combinations. In conclusion, the heterosyllabic clusters represents an asymmetric tendency with regard to place and manner features in English phonotactics: they tend to be similar violating OCP in terms of the place of articulation, but they tend to be different satisfying OCP in terms of the manner of articulation. We can also capture the effects are applied in gradient fashion. Additional finding of this study is that the OCP-voice and the Syllable Contact Law (Vennemann 1988) effects are not active in heterosyllabic clusters of English words.

To obtain Nuclear Magnetic Resonance (NMR) measurement of membrane protein, an NMR magnet is required to generate high intensity, homogeneity, and stability of field. A High-Temperature Superconducting (HTS) magnet is a promising alternative to a conventional Low-Temperature Superconducting (LTS) NMR magnet for high field, current density, and stability margin. Conventionally, an HTS coil has been wound by several winding techniques such as Single-Pancake (SP), Double-Pancake (DP), and layer-wound. The DP winding technique has been frequently used for a large magnet because long HTS wire is generally difficult to manufacture, and maintenance of magnet is convenient. However, magnetic field generated by the slanted turns and the splice leads to field inhomogeneity in Diameter of Spherical Volume (DSV). The field inhomogeneity degrades performance of NMR spectrometer and thus effect of the slanted turns and the splice should be analyzed. In this paper, field gradient of HTS double-pancake coils considering the slanted turns and the splice was calculated using Biot-Savart law and numerical integration. The calculation results showed that magnetic field produced by the slanted turns and the splice caused significant inhomogeneity of field.

A high gradient magnetic (HGM) separator prototype with the 2nd generation high temperature superconducting (2G HTS) magnetic system operated in sub-cooled nitrogen is presently under development at NRC “Kurchatov Institute” (Moscow, Russia). The main goal of the project is an attempt to shift away from the complicated liquid helium cryostats towards simple cryocooler-based nitrogen cryogenics as much more convenient for HGM separators industrial applications. Using of commercial HTS tapes allows to get a sufficient level of magnetic fields and extraction forces with low energy consumption. The expected operational parameters of the device are 1.2-1.5 T in the empty operational gap and up to 3 T on the ferromagnetic filters. In this paper we briefly describe the design of the HTS rotary separator prototype with the horizontally oriented rotor axis and propose different types of ferromagnetic filters intended for weakly magnetic ores enrichment.

In this study, we investigated high-gradient magnetic separation as a method for separating crud in high-temperature, high-pressure water inside a nuclear reactor. Corrosion products in the coolant circulate through the system and attach to the reactor core, where they are activated by neutron irradiation. The activated corrosion products then desorb from the core and circulate through the cooling system again. The corrosion product in the reactor water or piping system is called crud. Crud is the main source of radiation exposure for radiation workers. Removal and recovery of crud is important in decommissioning plants that have been in operation for service life, and new technologies are also desired. A method for separating activated ions adsorbed on ion exchange resins in nuclear reactors using magnetic separation is developed. In this method, the ion exchange resin is washed with acid, the activated ions are adsorbed from the washing water using adsorbents, and then separated magnetically. Rudimentary experiments were conducted to investigate the possibility of this method.

The purpose of this paper is to investigate gradience of identity avoidance in terms of speakers’ variation. In a Korean total reduplication where a consonant is epenthesized in the initial onset of the reduplicants, a general tendency was found: the epenthetic consonants incline to be dissimilar from the contextual consonants. This tendency, however, is not an across-the-board phenomenon. One of the factors for this non-categoricalness was found to be variation among speakers. A series of experiments revealed that different speakers showed their own preferences among different consonants for insertion. It was further found that among different groups of speakers due to their preferences for inserted consonants, there was also micro-level variation. Some speakers sticked to the same preferred segment for consonant insertion, whereas others chose another consonant when they were faced with their own preferred consonant in the existing context. This latter group of speakers seem to be affected by the principle of identity avoidance, albeit their preference for a certain consonant.

The manufacturing process of antibody drugs comprises two main stages: the upstream process for antibody cultivation and the downstream process for antibody extraction. The domestic bio industry has excellent technology for the upstream process. However, it relies on the technology of foreign countries to execute downstream process such as affinity chromatography. Furthermore, there are no domestic companies capable of producing the equipment for affinity chromatography. High gradient magnetic separation technology using a high temperature superconducting magnet as a novel antibody separation and purification technology is introduced to substitute for the traditional technology of affinity chromatography. A specially designed magnetic filter was equipped in the bore of the superconducting magnet enabling the continuous magnetic separation of nano-sized paramagnetic beads that can be used as affinity magnetic nano beads for antibodies. To optimize the magnetic filter that captures superparamagnetic nanoparticles effectively, various shapes and materials were examined for the magnetic filter. The result of magnetic separation experiments show that the maximum separation and recovery ratio of superparamagnetic nanoparticles are 99.2 %, and 99.07 %, respectively under magnetic field (3 T) and flow rate (600 litter/hr).

This present study was conducted to investigate protective effect of discontinuous Percoll gradient containing alpha-linolenic acid (ALA) before freezing process on viability, acrosome damage, mitochondrial activity, and oxidative stress of frozen-thawed boar spermatozoa. The separation of spermatozoa by discontinuous Percoll gradient was performed by different concentration of Percoll solution (45/90%) containing ALA combined with bovine serum albumin (BSA), and collected sperm in each Percoll layer was cryopreserved. To evaluate viability, acrosome damage, mitochondrial activity, and reactive oxygen species (ROS) level of frozen-thawed sperm, flow cytometry was used. Morphological abnormalities were observed under light microscope. In results, viability of sperm from 90% Percoll layer was higher than control and 45% Percoll group (p < 0.05). Separated sperm in 90% Percoll layer had lower acrosome damage and morphological abnormalities than control as well as viability, whereas 45% Percoll group was higher (p < 0.05). Similar with acrosome damage and abnormalities, mitochondrial activity was slightly enhanced and the population of live sperm with high ROS level was decreased by 90% Percoll separation, however, there was no significant difference. Supplementation of 3 ng/mL ALA into Percoll solution increased sperm viability and decreased population of live sperm with high ROS compared to control (p < 0.05). In conclusion, discontinuous Percoll gradient before freezing process could improve efficiency of cryopreservation of boar sperm through selection of sperm with high freezing resistance, and supplement of ALA during Percoll gradient might contribute suppression of ROS generation via stabilizing of plasma membrane during cryopreservation.