• ETRI Journal
• /
• 제41권6호
• /
• pp.760-770
• /
• 2019

Based on the growing demand for neural network technologies, various neural network inference engines are being developed. However, each inference engine has its own neural network storage format. There is a growing demand for standardization to solve this problem. This study presents interworking techniques for ensuring the compatibility of neural networks and data among the various deep learning frameworks. The proposed technique standardizes the graphic expression grammar and learning data storage format using the Neural Network Exchange Format (NNEF) of Khronos. The proposed converter includes a lexical, syntax, and parser. This NNEF parser converts neural network information into a parsing tree and quantizes data. To validate the proposed system, we verified that MNIST is immediately executed by importing AlexNet's neural network and learned data. Therefore, this study contributes an efficient design technique for a converter that can execute a neural network and learned data in various frameworks regardless of the storage format of each framework.

• ETRI Journal
• /
• 제45권2호
• /
• pp.318-328
• /
• 2023

Recently, embedded systems, such as mobile platforms, have multiple processing units that can operate in parallel, such as centralized processing units (CPUs) and neural processing units (NPUs). We can use deep-learning compilers to generate machine code optimized for these embedded systems from a deep neural network (DNN). However, the deep-learning compilers proposed so far generate codes that sequentially execute DNN operators on a single processing unit or parallel codes for graphic processing units (GPUs). In this study, we propose PartitionTuner, an operator scheduler for deep-learning compilers that supports multiple heterogeneous PUs including CPUs and NPUs. PartitionTuner can generate an operator-scheduling plan that uses all available PUs simultaneously to minimize overall DNN inference time. Operator scheduling is based on the analysis of DNN architecture and the performance profiles of individual and group operators measured on heterogeneous processing units. By the experiments for seven DNNs, PartitionTuner generates scheduling plans that perform 5.03% better than a static type-based operator-scheduling technique for SqueezeNet. In addition, PartitionTuner outperforms recent profiling-based operator-scheduling techniques for ResNet50, ResNet18, and SqueezeNet by 7.18%, 5.36%, and 2.73%, respectively.

• 한국정보통신학회논문지
• /
• 제27권1호
• /
• pp.34-40
• /
• 2023

고객이 온라인으로 요청한 명함을 자동으로 명함을 인쇄하는 스마트 명함 인쇄 시스템이 활성화되고 있다. 이때, 문제는 고객이 시스템에 제출한 명함이 비정상일 수 있다는 것이다. 본 논문에서는 인공 지능 기술을 도입하여 명함의 이미지가 비정상적으로 회전됐는지 여부를 판정하는 문제를 다룬다. 명함은 0도, 90도, 180도, 270도 회전한다고 가정하였다. 특별한 인공신경망을 설계하지 않고 기존의 VGG, ResNet, DenseNet 인공신경망을 적용하여 실험하였는데 모든 신경망이 97% 정도의 정확도로 이미지 회전을 분별할 수 있었다. DenseNet161은 97.9%의 정확도를 보였고 ResNet34도 97.2%의 정밀도를 보였다. 이는 문제가 단순할 경우, 복잡한 인공신경망이 아니어도 충분히 좋은 결과를 낼 수 있음을 시사한다.

• 융합보안논문지
• /
• 제23권1호
• /
• pp.27-33
• /
• 2023

Covid-19으로 인한 디지털화는 인공지능 기반의 음성인식 기술을 급속하게 발전시켰다. 그러나 이 기술은 데이터셋이 일부 집단에 편향될 경우 인종 및 성차별과 같은 불공정한 사회적 문제를 초래하고 인공지능 서비스의 신뢰성과 보안성을 열화시키는 요인이 된다. 본 연구에서는 대표적인 인공지능의 CNN(Convolutional Neural Network) 모델인 VGGNet(Visual Geometry Group Network), ResNet(Residual neural Network), MobileNet을 활용한 편향된 데이터 환경에서 정확도에 기반한 불공정성을 비교 및 분석한다. 실험 결과에 따르면 Top1-accuracy에서 ResNet34가 여성과 남성이 91%, 89.9%로 가장 높은 정확도를 보였고, 성별 간 정확도 차는 ResNet18이 1.8%로 가장 작았다. 모델별 성별 간의 정확도 차이는 서비스 이용 시 남녀 간의 서비스 품질에 대한 차이와 불공정한 결과를 야기한다.

• 한국유체기계학회 논문집
• /
• 제13권6호
• /
• pp.19-24
• /
• 2010

The demands for small, high performance and high loaded aircraft compressor are increased in the world. But the design requirements become increasingly complex to design these high technical engines, the requirement of the design optimization become increased. The optimal design result of several disciplines show different tendencies and nonlinear characteristics of the compressor design, the multidisciplinary design optimization method must be considered in compressor design. Therefore, the artificial Neural Net method is adapted to make the approximation model of 3-stage axial compressor design optimization for considering the nonlinear characteristic. At last, the optimal result of this study is compared to that of previous study.

• 한국환경보건학회지
• /
• 제28권3호
• /
• pp.19-25
• /
• 2002

In this paper, the water quality forecast was performed on the BOD of the Chungju Dam using the ARIMA model, which is a nonlinear statistics model, and the artificial neural network model. The monthly data of water quality were collected from 1991 to 2000. The most appropriate ARIMA model for Chungju dam was found to be the multiplicative seasonal ARIMA(1,0,1)(1,0,1)$_{12}$, model. While the artificial neural network model, which is used relatively often in recent days, forecasts new data by the strength of a learned matrix like human neurons. The BOD values were forecasted using the back-propagation algorithm of multi-layer perceptrons in this paper. Artificial neural network model was com- posed of two hidden layers and the node number of each hidden layer was designed fifteen. It was demonstrated that the ARIMA model was more appropriate in terms of changes around the overall average, but the artificial neural net-work model was more appropriate in terms of reflecting the minimum and the maximum values.s.

• Nuclear Engineering and Technology
• /
• 제54권12호
• /
• pp.4776-4797
• /
• 2022

In the one-dimensional thermal-hydraulic (TH) codes, a subcooled boiling model to predict the void fraction profiles in a vertical channel consists of wall heat flux partitioning, the vapor condensation rate, the bubbly-to-slug flow transition criterion, and drift-flux models. Model performance has been investigated in detail, and necessary refinements have been incorporated into the Safety and Performance Analysis Code (SPACE) developed by the Korean nuclear industry for the safety analysis of pressurized water reactors (PWRs). The necessary refinements to models related to pumping factor, net vapor generation (NVG), vapor condensation, and drift-flux velocity were investigated in this study. In particular, a new NVG empirical correlation was also developed using artificial neural network (ANN) techniques. Simulations of a series of subcooled flow boiling experiments at pressures ranging from 1 to 149.9 bar were performed with the refined SPACE code, and reasonable agreement with the experimental data for the void fraction in the vertical channel was obtained. From the root-mean-square (RMS) error analysis for the predicted void fraction in the subcooled boiling region, the results with the refined SPACE code produce the best predictions for the entire pressure range compared to those using the original SPACE and RELAP5 codes.

• 한국음향학회지
• /
• 제39권5호
• /
• pp.454-460
• /
• 2020

음성인식 기능을 제공하는 인공지능 비서들은 정확도가 뛰어난 클라우드 기반의 음성인식을 통해 동작한다. 클라우드 기반의 음성인식에서 시작 단어 인식은 대기 중인 기기를 활성화하는 데 중요한 역할을 한다. 본 논문에서는 공개 데이터셋인 구글의 Speech Commands 데이터셋을 사용하여 스펙트로그램 및 멜-주파수 캡스트럼 계수 특징을 입력으로 하여 모바일 기기에 대응한 저 연산 시작 단어 검출을 위한 합성곱 신경망의 성능을 비교한다. 본 논문에서 사용한 합성곱 신경망은 다층 퍼셉트론, 일반적인 합성곱 신경망, VGG16, VGG19, ResNet50, ResNet101, ResNet152, MobileNet이며, MobileNet의 성능을 유지하면서 모델 크기를 1/25로 줄인 네트워크도 제안한다.

• 한국구조물진단유지관리공학회 논문집
• /
• 제2권4호
• /
• pp.137-147
• /
• 1998

The neural net application was tried to develop the technique for monitoring the health status of a steel truss bridge which was scaled down to 1/15 of the real bridge for the laboratory experiments. The damage scenarios were chosen as 7 cases. The dynamic behavior, which was changed due to the breakage of the members, of the bridge was investigated by finite element analysis. The bridge consists of single spam, and eight (8) main structural subsystems. The loading vehicle, which weighs as 100 kgf, was operated by the servo-motor controller. The accelerometers were bonded on the surface of 7 cross-beams to measure the dynamic behavior induced by the abnormal structural condition. Artificial neural network technique was used to determine the severity of the damage. At first, the neural net was learnt by the results of finite element analysis, and also, the maximum detection error was 3.65 percents. Another neural net was also learnt, and verified by the experimental results, and in this case, the maximum detection error was 1.05 percents. In future study, neural net is necessary to be learnt and verified by various data from the real bridge.

• 품질경영학회지
• /
• 제49권4호
• /
• pp.539-550
• /
• 2021

Purpose: The purpose of this study is to propose a multivariate CUSUM control chart that can detect the out-of-control state fast while monitoring the cross- and auto- correlated multivariate time series data. Methods: We first build models to estimate the observation data and calculate the corresponding residuals. After then, a multivariate CUSUM chart is applied to monitor the residuals instead of the original raw observation data. Vector Autoregression and Artificial Neural Net are selected for the modelling, and Separated-MCUSUM chart is selected for the monitoring. The suggested methods are tested under a number of experimental settings and the performances are compared with those of other existing methods. Results: We find that Artificial Neural Net is more appropriate than Vector Autoregression for the modelling and show the combination of Separated-MCUSUM with Artificial Neural Net outperforms the other alternatives considered in this paper. Conclusion: The suggested chart has many advantages. It can monitor the complicated multivariate data with cross- and auto- correlation, and detects the out-of-control state fast. Unlike other CUSUM charts finding their control limits by trial and error simulation, the suggested chart saves lots of time and effort by approximating its control limit mathematically. We expect that the suggested chart performs not only effectively but also efficiently for monitoring the process with complicated correlations and frequently-changed parameters.