• 융합보안논문지
• /
• 제22권4호
• /
• pp.129-133
• /
• 2022

딥러닝 프레임워크는 현재에도 계속해서 발전되어 가고 있으며, 다양한 프레임워크들이 존재한다. 딥러닝의 대표적인 프레임워크는 TensorFlow, PyTorch, Keras 등이 있다. 딥러님 프레임워크는 이미지 학습을 통해 이미지 분류에서의 최적화 모델을 이용한다. 본 논문에서는 딥러닝 이미지 인식 분야에서 가장 많이 사용하고 있는 TensorFlow와 PyTorch 프레임워크를 활용하여 이미지 학습을 진행하였으며, 이 과정에서 도출한 결과를 비교 분석하여 최적화된 프레임워크을 알 수 있었다.

• 디지털산업정보학회논문지
• /
• 제19권3호
• /
• pp.245-251
• /
• 2023

Facial expression recognition plays a significant role in understanding human emotional states. With the advancement of AI and computer vision technologies, extensive research has been conducted in various fields, including improving customer service, medical diagnosis, and assessing learners' understanding in education. In this study, we develop a model that can infer emotions in real-time from a webcam using transfer learning with TensorFlow.js and MobileNet. While existing studies focus on achieving high accuracy using deep learning models, these models often require substantial resources due to their complex structure and computational demands. Consequently, there is a growing interest in developing lightweight deep learning models and transfer learning methods for restricted environments such as web browsers and edge devices. By employing MobileNet as the base model and performing transfer learning, our study develops a deep learning transfer model utilizing JavaScript-based TensorFlow.js, which can predict emotions in real-time using facial input from a webcam. This transfer model provides a foundation for implementing facial expression recognition in resource-constrained environments such as web and mobile applications, enabling its application in various industries.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2017년도 추계학술발표대회
• /
• pp.386-388
• /
• 2017

TensorFlow와 알파고의 등장으로 인공지능의 높은 성능과 다양한 활용 가능성을 보이면서, 폭 넓은 산업 분야에서 머신러닝 기술에 대한 수요가 증가하고 있다. 반면, 머신러닝 기술은 GPU 기반 고속 병렬처리 기술과 인프라 기술을 기반으로 하고 있기 때문에, 머신러닝 기반 서비스 개발 및 제공에 어려움을 겪고 있다. 본 논문에서는 이와 같은 문제를 개선하기 위해서 개발한 고성능 GPU 기반 컨테이너 클라우드 시스템을 소개한다. 해당 시스템은 GPU 기반 고속 병렬처리를 지원하고, Kubernetes 클러스터에서 컨테이너를 기반으로 TensorFlow Serving을 손쉽게 배포할 수 있는 기능을 제공한다.

• 한국복합재료학회:학술대회논문집
• /
• 한국복합재료학회 2005년도 춘계학술발표대회 논문집
• /
• pp.134-136
• /
• 2005

Homogenization method is adopted to predict the permeability tenor for glass fiber plain woven fabrics. Calculating the permeability tensor numerically is an encouraging task because the permeability tensor is a key parameter in resin transfer molding (RTM). Based on multi-scale approach of the homogenization method, the permeability for the micro-unit cell within fiber tow is computed and compared with that obtained from flow analysis for the same micro-unit cell. It is found that they are in good agreement. In order to calculate the permeability tensor of macro-unit cell for the plain woven fabrics, the Stokes and Brinkman equations which describe inter-tow and intra-tow flow respectively are employed as governing equations. The effective permeabilities homogenized by considering intra-tow flow are compared with those obtained experimentally. Control volume finite element method (CVFEM) is used as a numerical method. It is shown that the asymptotic expansion homogenization method is an attractive method to predict the effective permeability for heterogeneous media.

• International Journal of Computer Science & Network Security
• /
• 제22권10호
• /
• pp.237-245
• /
• 2022

The prominence of IoTs (Internet of Things) and exponential advancement of computer networks has resulted in massive essential applications. Recognizing various cyber-attacks or anomalies in networks and establishing effective intrusion recognition systems are becoming increasingly vital to current security. MLTs (Machine Learning Techniques) can be developed for such data-driven intelligent recognition systems. Researchers have employed a TFDNNs (Tensor Flow Deep Neural Networks) and DCNNs (Deep Convolution Neural Networks) to recognize pirated software and malwares efficiently. However, tuning the amount of neurons in multiple layers with activation functions leads to learning error rates, degrading classifier's reliability. HTFDNNs ( Hybrid tensor flow DNNs) and MRNs (Modified Residual Networks) or Resnet CNNs were presented to recognize software piracy and malwares. This study proposes HTFDNNs to identify stolen software starting with plagiarized source codes. This work uses Tokens and weights for filtering noises while focusing on token's for identifying source code thefts. DLTs (Deep learning techniques) are then used to detect plagiarized sources. Data from Google Code Jam is used for finding software piracy. MRNs visualize colour images for identifying harms in networks using IoTs. Malware samples of Maling dataset is used for tests in this work.

• Korea-Australia Rheology Journal
• /
• 제16권4호
• /
• pp.183-191
• /
• 2004

High Deborah or Weissenberg number problems in viscoelastic flow modeling have been known formidably difficult even in the inertialess limit. There exists almost no result that shows satisfactory accuracy and proper mesh convergence at the same time. However recently, quite a breakthrough seems to have been made in this field of computational rheology. So called matrix-logarithm (here we name it tensor-logarithm) formulation of the viscoelastic constitutive equations originally written in terms of the conformation tensor has been suggested by Fattal and Kupferman (2004) and its finite element implementation has been first presented by Hulsen (2004). Both the works have reported almost unbounded convergence limit in solving two benchmark problems. This new formulation incorporates proper polynomial interpolations of the log­arithm for the variables that exhibit steep exponential dependence near stagnation points, and it also strictly preserves the positive definiteness of the conformation tensor. In this study, we present an alternative pro­cedure for deriving the tensor-logarithmic representation of the differential constitutive equations and pro­vide a numerical example with the Leonov model in 4:1 planar contraction flows. Dramatic improvement of the computational algorithm with stable convergence has been demonstrated and it seems that there exists appropriate mesh convergence even though this conclusion requires further study. It is thought that this new formalism will work only for a few differential constitutive equations proven globally stable. Thus the math­ematical stability criteria perhaps play an important role on the choice and development of the suitable con­stitutive equations. In this respect, the Leonov viscoelastic model is quite feasible and becomes more essential since it has been proven globally stable and it offers the simplest form in the tensor-logarithmic formulation.

• 한국항행학회논문지
• /
• 제21권6호
• /
• pp.674-679
• /
• 2017

4차 산업혁명과 ICT 기술의 중요성이 증가함에 따라 소프트웨어 중심 사회가 초래되었다. 기존 소프트웨어 교육은 학습 환경구성에 제한적이었으며, 초기에 많은 비용이 발생하였다. 이를 해결하기 위하여 웹 컴파일러를 활용하는 형태의 학습 방법이 개발되었다. 웹 컴파일러는 다양한 소프트웨어 언어를 지원하며, 컴파일 결과를 사용자에게 웹을 통해 보여준다. 하지만 4차 산업혁명의 핵심기술인 인공지능에 대한 웹 컴파일러는 아직 미비한 상황이다. 본 논문에서는 구글 인공지능 라이브러리인 텐서플로우 기반의 웹 컴파일러를 설계, 구현하였다. nodeJS 기반의 서버에 텐서플로우와 텐서플로우 서빙, 파이썬 주피터를 구현하고, meteorJS 기반의 웹 서버를 구축하여 인공지능 학습을 위한 시스템을 구현하였다. 소프트웨어 중심 사회에서 인공지능 학습을 위한 도구로써의 활용이 가능할 것으로 기대된다.

• Journal of Mechanical Science and Technology
• /
• 제16권11호
• /
• pp.1522-1539
• /
• 2002

In the present study, an explicit algebraic stress model is shown to be the exact tensor representation of algebraic stress model by directly solving a set of algebraic equations without resort to tensor representation theory. This repeals the constraints on the Reynolds stress, which are based on the principle of material frame indifference and positive semi-definiteness. An a priori test of the explicit algebraic stress model is carried out by using the DNS database for a fully developed channel flow at Rer = 135. It is confirmed that two-point correlation function between the velocity fluctuation and the Laplacians of the pressure-gradient i s anisotropic and asymmetric in the wall-normal direction. Thus, a novel composite algebraic Reynolds stress model is proposed and applied to the channel flow calculation, which incorporates non-local effect in the algebraic framework to predict near-wall behavior correctly.

• 대한기계학회논문집
• /
• 제18권4호
• /
• pp.967-976
• /
• 1994

A tensor invariant model equation for the turbulent energy dissipation rate is proposed in the present study, which is able to simulate secondary straining effects such as curvature effects without the introduction of additional empirical input. The source term in this model has a combined form of the generation term due to the mean vorticity with the conventional one due to the mean strain rate. An extended low-Reynolds-number $k-\epsilon$ turbulence model involving this new model equation is tested for a turbulent Coutte flow between coaxial cylinders with inner cylinder rotated, which is a well defined example of curved flows. The predicted results indicate that the present model works much better for this flow, compared with previous models.

• 유체기계공업학회:학술대회논문집
• /
• 유체기계공업학회 2003년도 유체기계 연구개발 발표회 논문집
• /
• pp.496-501
• /
• 2003

Reynolds averaged Wavier-Stokes simulations based on the Reynolds stress model was performed to investigated the effect of inlet flow angle on the distributions of the Reynolds stress tensor inside tip leakage vortex of a linear compressor cascade. Two different inlet flow angles ${\beta}=29.3^{\circ}$(design condition) and $36.5^{\circ}$(off-design condition) were considered. Stress tensor analysis, which transforms the Reynolds stress into the principal direction, was applied to show an anisotropy of the normal stresses. Whereas the anisotropy was highest in the region where the tip leakage vortex collides the suction side of the blade and tip leakage flow enters between blade tip of the pressure side and the endwall, it had the lowest value at the center of tip leakage vortex. It was also found that the magnitude of maximum shear stress at design condition was greater than that of off-design condition.