• 정보저장시스템학회논문집
• /
• 제6권1호
• /
• pp.18-24
• /
• 2010

The performance of the speech recognition mainly depends on uncertain factors such as speaker's conditions and environmental effects. The present study deals with the speech recognition of a number of multi-syllable isolated Korean words using soft computing techniques such as back-propagation neural network, fuzzy inference system, and fuzzy neural network. Feature patterns for the speech recognition are analyzed with 12th order thirty frames that are normalized by the linear predictive coding and Cepstrums. Using four models of speech recognizer, actual experiments for both single-speakers and multiple-speakers are conducted. Through this study, the recognizers of combined fuzzy logic and back-propagation neural network and fuzzy neural network show the better performance in identifying the speech recognition.

• 한국컴퓨터그래픽스학회논문지
• /
• 제26권3호
• /
• pp.87-97
• /
• 2020

본 논문에서는 사용자가 지정한 반사 하이라이트 맵을 가이드 영상으로 하는 뉴럴 재조명 기법을 제안한다. 제안하는 신경망은 다양한 조명 위치에서 렌더링 된 영상을 이용해 사전학습시킨 백본 뉴럴 렌더러를 활용하며, 기저 영상과 렌더링 영상의 차이가 사용자가 제공한 반사 하이라이트 맵과 유사하도록 역전파에 의해 광원의 위치와 관련된 재조명 영상을 동시 최적화한다. 제안하는 방법은 아티스트가 선호하는 이차원 화면 공간 인터페이스를 제공하면서도 삼차원 조명의 위치를 명시적으로 추론할 수 있는 장점이 있다. 제안하는 뉴럴 재조명의 성능은 실제 값을 설정할 수 있는 실험 상황을 수립하여, 본 논문의 방법이 주어진 하이라이트 맵을 얼마나 잘 반영하는지 평가하고 실제 하이라이트 맵으로 추론한 조명 및 재조명 영상의 오차를 측정하였다. 제안하는 뉴럴 재조명이 추정한 광원 위치의 평균 오차율은 정규화된 삼차원 장면 크기 대비 0.11이다.

• Nuclear Engineering and Technology
• /
• 제43권3호
• /
• pp.243-256
• /
• 2011

This paper considers the concept of analog computing based on a cellular neural network (CNN) paradigm to simulate nuclear reactor dynamics using a time-dependent second order form of the neutron transport equation. Instead of solving nuclear reactor dynamic equations numerically, which is time-consuming and suffers from such weaknesses as vulnerability to transient phenomena, accumulation of round-off errors and floating-point overflows, use is made of a new method based on a cellular neural network. The state-of-the-art shows the CNN as being an alternative solution to the conventional numerical computation method. Indeed CNN is an analog computing paradigm that performs ultra-fast calculations and provides accurate results. In this study use is made of the CNN model to simulate the space-time response of scalar flux distribution in steady state and transient conditions. The CNN model also is used to simulate step perturbation in the core. The accuracy and capability of the CNN model are examined in 2D Cartesian geometry for two fixed source problems, a mini-BWR assembly, and a TWIGL Seed/Blanket problem. We also use the CNN model concurrently for a typical small PWR assembly to simulate the effect of temperature feedback, poisons, and control rods on the scalar flux distribution.

• ETRI Journal
• /
• 제42권1호
• /
• pp.90-100
• /
• 2020

Named entity recognition (NER) continues to be an important task in natural language processing because it is featured as a subtask and/or subproblem in information extraction and machine translation. In Urdu language processing, it is a very difficult task. This paper proposes various deep recurrent neural network (DRNN) learning models with word embedding. Experimental results demonstrate that they improve upon current state-of-the-art NER approaches for Urdu. The DRRN models evaluated include forward and bidirectional extensions of the long short-term memory and back propagation through time approaches. The proposed models consider both language-dependent features, such as part-of-speech tags, and language-independent features, such as the "context windows" of words. The effectiveness of the DRNN models with word embedding for NER in Urdu is demonstrated using three datasets. The results reveal that the proposed approach significantly outperforms previous conditional random field and artificial neural network approaches. The best f-measure values achieved on the three benchmark datasets using the proposed deep learning approaches are 81.1%, 79.94%, and 63.21%, respectively.

• 예술인문사회 융합 멀티미디어 논문지
• /
• 제7권11호
• /
• pp.71-79
• /
• 2017

본 논문에서는 이기종 컴퓨팅을 활용한 환율 예측 뉴럴 네트워크를 구현했다. 환율 예측에는 많은 양의 데이터가 필요하다. 그에 따라 이러한 데이터를 활용할 수 있는 뉴럴 네트워크를 사용했다. 뉴럴 네트워크는 크게 학습과 검증의 두 과정을 거친다. 학습은 CPU를 활용했다. 검증에는 Verilog HDL로 작성된 RTL을 FPGA에서 동작 시켰다. 해당 뉴럴 네트워크의 구조는 입력 뉴런 네 개, 히든 뉴런 네 개, 출력 뉴런 한 개를 가진다. 입력 뉴런에는 미국 1달러, 일본 100엔, EU 1유로, 영국 1파운드의 원화 가치를 사용했다. 입력 뉴런들을 통해 캐나다 1달러의 원화가치를 예측 했다. 환율을 예측 하는 순서는 입력, 정규화, 고정 소수점 변환, 뉴럴 네트워크 순방향, 부동 소수점 변환, 역정규화, 출력 과정을 거친다. 2016년 11월의 환율을 예측한 결과 0.9원에서 9.13원 사이의 오차 금액이 발생했다. 환율 이외의 다른 데이터를 추가해 뉴런의 개수를 늘린다면 더 정확한 환율 예측이 가능할 것으로 예상된다.

• 한국전산구조공학회:학술대회논문집
• /
• 한국전산구조공학회 2000년도 봄 학술발표회논문집
• /
• pp.449-457
• /
• 2000

The paper describes the study of global approximate optimization utilizing soft computing techniques such as genetic algorithms (GA's), neural networks (NN's), and fuzzy inference systems(FIS). GA's provide the increasing probability of locating a global optimum over the entire design space associated with multimodality and nonlinearity. NN's can be used as a tool for function approximations, a rapid reanalysis model for subsequent use in design optimization. FIS facilitates to handle the quantitative design information under the case where the training data samples are not sufficiently provided or uncertain information is included in design modeling. Properties of soft computing techniques affect the quality of global approximate model. Evolutionary fuzzy modeling (EFM) and adaptive neuro-fuzzy inference system (ANFIS) are briefly introduced for structural optimization problem in this context. The paper presents the success of EFM depends on how optimally the fuzzy membership parameters are selected and how fuzzy rules are generated.

• Advances in nano research
• /
• 제15권6호
• /
• pp.513-520
• /
• 2023

The exponential proliferation of cutting-edge computing technologies has spurred organizations to outsource their data and computational needs. In the realm of cloud-based computing environments, ensuring robust security, encompassing principles such as confidentiality, availability, and integrity, stands as an overarching imperative. Elevating security measures beyond conventional strategies hinges on a profound comprehension of malware's multifaceted behavioral landscape. This paper presents an innovative paradigm aimed at empowering cloud service providers to adeptly model user behaviors. Our approach harnesses the power of a Particle Swarm Optimization-based Probabilistic Neural Network (PSO-PNN) for detection and recognition processes. Within the initial recognition module, user behaviors are translated into a comprehensible format, and the identification of malicious nano-structures behaviors is orchestrated through a multi-layer neural network. Leveraging the UNSW-NB15 dataset, we meticulously validate our approach, effectively characterizing diverse manifestations of malicious nano-structures behaviors exhibited by users. The experimental results unequivocally underscore the promise of our method in fortifying security monitoring and the discernment of malicious nano-structures behaviors.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 2001년도 추계학술대회 학술발표 논문집
• /
• pp.133-136
• /
• 2001

This paper partially presents a Hierachical neural network architecture for providing the intelligent control of complex Automatic Transmission(AJT) system which is usually nonlinear and hard to model mathematically. It consists of the module to apply or release an engine brake at the slope and that to judge the intention of the driver. The HNN architecture simplifies the structure of the overall system and is efficient for the learning time. This paper describes how the sub-neural networks of each module have been constructed and will compare the result of the intelligent hJT control to that of the conventional shift pattern.

• 전자통신동향분석
• /
• 제35권4호
• /
• pp.34-41
• /
• 2020

The existing Von Neumann architecture places limits to data processing in AI, a booming technology. To address this issue, research is being conducted on computing architectures and artificial neural networks that simulate neurons and synapses, which are the hardware of the human brain. With high-speed, high-throughput data communication infrastructures, photonic solutions today are a mature industrial reality. In particular, due to the recent outstanding achievements of artificial neural networks, there is considerable interest in improving their speed and energy efficiency by exploiting photonic-based neuromorphic hardware instead of electronic-based hardware. This paper covers recent photonic neuromorphic studies and a classification of existing solutions (categorized into multilayer perceptrons, convolutional neural networks, spiking neural networks, and reservoir computing).

• ETRI Journal
• /
• 제35권2호
• /
• pp.207-217
• /
• 2013

This paper introduces the modular cellular neural network (CNN), which is a new CNN structure constructed from nine one-layer modules with intercellular interactions between different modules. The new network is suitable for implementing many image processing operations. Inputting an image into the modules results in nine outputs. The topographic characteristic of the cell interactions allows the outputs to introduce new properties for image processing tasks. The stability of the system is proven and the performance is evaluated in several image processing applications. Experiment results on texture segmentation show the power of the proposed structure. The performance of the structure in a real edge detection application using the Berkeley dataset BSDS300 is also evaluated.