• 한국음향학회지
• /
• 제41권1호
• /
• pp.30-37
• /
• 2022

본 논문은 주파수 영역에서 심층 신경망 기반 음성 향상 모델 학습을 위하여 학습 대상과 네트워크 구조에 따라 두 가지 관점에서 성능을 비교 평가한다. 이때, 학습 대상으로는 스펙트럼 매핑과 Time-Frequency(T-F) 마스킹 기법을 사용하였고 네트워크 구조는 실수 네트워크와 복소 네트워크를 사용하였다. 음성 향상 모델의 성능은 데이터 셋 규모에 따라 Perceptual Evaluation of Speech Quality(PESQ)와 Short-Time Objective Intelligibility(STOI) 두 가지 객관적 평가지표를 통해 평가하였다. 실험 결과, 네트워크의 종류와 데이터 셋 종류에 따라 적정한 훈련 데이터의 크기가 다르다는 것을 확인하였다. 또한, 데이터의 크기와 학습 대상에 따라 복소 네트워크보다 실수 네트워크가 비교적 높은 성능을 보이기 때문에 총 파라미터의 수를 고려한다면 경우에 따라 실수 네트워크를 사용하는 것이 보다 현실적인 해결책일 수 있다는 것을 확인하였다.

• Structural Engineering and Mechanics
• /
• 제45권1호
• /
• pp.1-18
• /
• 2013

The typhoon wind characteristics designing for buildings or bridges located in complex terrain and typhoon prone region normally cannot be achieved by the very often few field measurement data, or by physical simulation in wind tunnel. This study proposes a numerical simulation procedure for predicting directional typhoon design wind speeds and profiles for sites over complex terrain by integrating typhoon wind field model, Monte Carlo simulation technique, CFD simulation and artificial neural networks (ANN). The site of Stonecutters Bridge in Hong Kong is chosen as a case study to examine the feasibility of the proposed numerical simulation procedure. Directional typhoon wind fields on the upstream of complex terrain are first generated by using typhoon wind field model together with Monte Carlo simulation method. Then, ANN for predicting directional typhoon wind field at the site are trained using representative directional typhoon wind fields for upstream and these at the site obtained from CFD simulation. Finally, based on the trained ANN model, thousands of directional typhoon wind fields for the site can be generated, and the directional design wind speeds by using extreme wind speed analysis and the directional averaged mean wind profiles can be produced for the site. The case study demonstrated that the proposed procedure is feasible and applicable, and that the effects of complex terrain on design typhoon wind speeds and wind profiles are significant.

• Structural Engineering and Mechanics
• /
• 제16권5호
• /
• pp.581-595
• /
• 2003

Since the conventional direct approaches are hard to be applied for damage diagnosis of complex large-scale structures, a two-step approach for diagnosing the joint damage of framed structures is presented in this paper by using artificial neural networks. The first step is to judge the damaged areas of a structure, which is divided into several sub-areas, using probabilistic neural networks with natural Frequencies Shift Ratio inputs. The next step is to diagnose the exact damage locations and extents by using the Radial Basis Function (RBF) neural network with the second Element End Strain Mode of the damaged sub-area input. The results of numerical simulation show that the proposed approach could diagnose the joint damage of framed structures induced by earthquake action effectively and has reliable anti-jamming abilities.

• 한국콘크리트학회:학술대회논문집
• /
• 한국콘크리트학회 2002년도 봄 학술발표회 논문집
• /
• pp.545-550
• /
• 2002

Concrete is one of the essential structural materials in the construction. But, concrete consists of many materials and is affected by many factors such as properties of materials, site environmental situations, and skill of constructor. Therefore, concrete mixes depend on experiences of experts. However, it is more and more difficult to determine concrete mixes design by empirical means because more ingredients like mineral and chemical admixtures are included. Artificial Neural Networks(ANN) are a mimic models of human brain to solve a complex nonlinear problem. They are powerful pattern recognizers and classifiers, also their computing abilities have been proven in the fields of prediction, estimation and pattern recognition. Here, among them, the back propagation network and radial basis function network are used. Compositions of high-performance concrete mixes are eight components(water, cement, fine aggregate, coarse aggregate, fly ash, silica fume, superplasticizer and air-entrainer). Compressive strength and slump are measured. The results show that neural networks are proper tools to minimize the uncertainties of the design of concrete mixtures.

• BMB Reports
• /
• 제37권1호
• /
• pp.53-58
• /
• 2004

With advances in determining the entire DNA sequence of the human genome, it is now critical to systematically identify the function of a number of genes in the human genome. These biological challenges, especially those in human diseases, should be addressed in human cells in which conventional (e.g. genetic) approaches have been extremely difficult to implement. To overcome this, several approaches have been initiated. This review will focus on the development of a novel 'chemical genetic/genomic approach' that uses small molecules to 'probe and identify' the function of genes in specific biological processes or pathways in human cells. Due to the close relationship of small molecules with drugs, these systematic and integrative studies will lead to the 'medicinal systems biology approach' which is critical to 'formulate and modulate' complex biological (disease) networks by small molecules (drugs) in human bio-systems.

• Smart Structures and Systems
• /
• 제6권2호
• /
• pp.147-165
• /
• 2010

Impact damage detection in composite structures has gained a considerable interest in many engineering areas. The capability to detect damage at the early stages reduces any risk of catastrophic failure. This paper compares two advanced signal processing methods for impact location in composite aircraft structures. The first method is based on a modified triangulation procedure and Genetic Algorithms whereas the second technique applies Artificial Neural Networks. A series of impacts is performed experimentally on a composite aircraft wing-box structure instrumented with low-profile, bonded piezoceramic sensors. The strain data are used for learning in the Neural Network approach. The triangulation procedure utilises the same data to establish impact velocities for various angles of strain wave propagation. The study demonstrates that both approaches are capable of good impact location estimates in this complex structure.

• 한국기술혁신학회:학술대회논문집
• /
• 한국기술혁신학회 2003년도 춘계학술대회
• /
• pp.313-335
• /
• 2003

It has been thirty years since Daedeuk Science Park was established in Daejeon, Korea. The Park has grown to be a national center for scientific activities, hosting more thirty research institutes at major scale with 12,000 scientists. As a main purpose of establishing the science park was to promote sciences as basic research, there have not been much economic impacts created from the science park. There have also been very little networks established among research and educational institutes within the science park and the city of Daejeon in general. After the economic crisis in 1997, however, there have been increasing changes around the Park. Since then more than 400 technology-intensive firms were newly created, mainly from the research institutes, or introduced to the Park area. This paper analyzes R&D networks of these small firms, especially with research institutes and universities, based on a questionnaire survey. The paper has argued: 1) to establish technology transfer centers, 2) to emphasize research related to commercialization, 3) to intensify advertizement of research outputs to the potential entrepreneurs, 4) to assist entrepreneural associations of concerned companies.

• 대한전기학회논문지
• /
• 제40권8호
• /
• pp.816-824
• /
• 1991

It is well known that dynamic control is needed for fast and accurate control. Neural networks are ideal for representing the strongly nonlinear relationship in the dynamic equations including complex unmodeled effects. It thus creates many advantages over conventional methods such as simple, fast and accurate control through neural network's inherent learning and massive parallelism. In this paper, dynamic control of the full six degrees of freedom of an industrial robot arm will be presented using neural networks. Moreover, through application to a real robot the usefulness of neurocontrol is demonstrated. The back propagation and feedback-error learning is used to train the neurocontroller. Simulated control of a PUMA 560 arm demonstrates that it moves at high speed with good accuracy and generalizes over untrained trajectories as well as adapt to unforseen load changes and sensor noise.

• 한국지능시스템학회:학술대회논문집
• /
• 한국퍼지및지능시스템학회 1997년도 추계학술대회 학술발표 논문집
• /
• pp.413-416
• /
• 1997

Recently, dynamic recurrent neural networks(DRNN) for identification of nonlinear dynamic systems have been researched extensively. In general, dynamic backpropagation was used to adjust the weights of neural networks. But, this method requires many complex calculations and has the possibility of falling into a local minimum. So, we propose a new approach to identify nonlinear dynamic systems using DRNN. In order to adjust the weights of neurons, we use evolution strategies, which is a method used to solve an optimal problem having many local minimums. DRNN trained by evolution strategies with mutation as the main operator can act as a plant emulator. And the fitness function of evolution strategies is based on the difference of the plant's outputs and DRNN's outputs. Thus, this new approach at identifying nonlinear dynamic system, when applied to the simulation of a two-link robot manipulator, demonstrates the performance and efficiency of this proposed approach.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1993년도 정기총회 및 추계학술대회 논문집 학회본부
• /
• pp.142-144
• /
• 1993

In this study, a Self-learning Neural-Fuzzy Networks is presented, Because of the fuzzy controller property, the designing problems of fuzzy if-then rules, membership functions and inference methods are very complex task. Thus in this paper we proposed the Neural-Fuzzy Networks composed by Sugeno and Takagi's fuzzy inference method and learned by using temporal back propagation algorithm. The proposed method can refine automatically the fuzzy if-then rules without human expert's knowledges. The induction motor servo system is used to demonstrate the effectiveness of the proposed control scheme and the feasibility of the acquired fuzzy controller. All results are supported by simulation.