• 전자공학회논문지B
• /
• 제32B권10호
• /
• pp.1347-1357
• /
• 1995

In this paper, we describe a modularized neuroclassifier for enhancing the recognition accuracy of mixed printed and handwritten numerals. This classifier combines four modularized subclassifiers using multi-layer perceptron module. The input of each subclassifier is comprised of a group of specialized feature sets. On applying this method to combining several subclassifiers for unconstrained handwritten numerals, the experimental result shows that the performance of individual subclassifier can be improved. In winner-take-all voting method, the result of subclassifier having the highest RF value is selected as the output. The generality of this classifier is tested with 1,080 printed and 3,000 handwritten numerals that was not shown in training the neural networks. Experimental results show 98.2% recognition rate. The typical recognition test with a threshold value(RF=1.5) has shown 97% recognition, 1% substitution and 2% rejection rates.

• 산업경영시스템학회지
• /
• 제21권48호
• /
• pp.123-132
• /
• 1998

The group formation problem of the machine and part is a very important issue in the planning stage of cellular manufacturing systems. This paper investigates Self-Organizing Map(SOM) neural networks approach to machine-part grouping problem. We present a two-phase algorithm based on SOM for grouping parts and machines. SOM can learn from complex, multi-dimensional data and transform them into visually decipherable clusters. Output layer in SOM network is one-dimensional structure and the number of output node has been increased sufficiently to spread out the input vectors in the order of similarity. The proposed algorithm performs remarkably well in comparison with many other algorithms for the well-known problems shown in previous papers.

• ETRI Journal
• /
• 제40권5호
• /
• pp.643-652
• /
• 2018

Dysarthria is a degenerative disorder of the central nervous system that affects the control of articulation and pitch; therefore, it affects the uniqueness of sound produced by the speaker. Hence, dysarthric speaker recognition is a challenging task. In this paper, a feature-extraction method based on deep belief networks is presented for the task of identifying a speaker suffering from dysarthria. The effectiveness of the proposed method is demonstrated and compared with well-known Mel-frequency cepstral coefficient features. For classification purposes, the use of a multi-layer perceptron neural network is proposed with two structures. Our evaluations using the universal access speech database produced promising results and outperformed other baseline methods. In addition, speaker identification under both text-dependent and text-independent conditions are explored. The highest accuracy achieved using the proposed system is 97.3%.

• 전자공학회논문지B
• /
• 제30B권11호
• /
• pp.56-65
• /
• 1993

This paper presents the design consideration of the MFNNs(Multilayer Feed forward Neural Networks) based on the distribution of the given teching patterns. By extracting the feature points from the given teaching patterns, the structure of a network including the netowrk size and interconnection weights of a network is initialized. This network is trained based on the modified version of the EBP(Error Back Propagation) algorithm. As a result, the proposed method has the advantage of learning speed compared to the conventional learning of the MFNNs with randomly chosen initial weights. To show the effectiveness of the suggested approach, the simulation result on the approximation of a two demensional continuous function is shown.

• 한국항해항만학회:학술대회논문집
• /
• 한국항해항만학회 2017년도 추계학술대회
• /
• pp.50-53
• /
• 2017

장기 해운불황에 따라 불확실성이 증폭되고 있는 상황에서 경기추세의 이해뿐만 아니라 예측 또한 중요해지고 있는 실정이다. 본 논문에서는 최근 특정 복잡한 문제에 대해서 각광받고 있는 인공신경망을 적용하여 BDI 예측을 연구하였다. 사용된 인공신경망은 순환신경망으로 RNN과 LSTM 그리고 비교의 목적으로 MLP를 통해 2009.04.01.부터 2017.07.31.의 기간을 대상으로 연구를 진행하였다. 또한 전통적 시계열 예측방법론인 ARIMA 분석을 실시해 인공신경망들의 예측성능과 비교하였다. 결과로 순환신경망인 RNN의 성능이 가장 뛰어났으며 LSTM의 특정 시계열(BDI)에의 적용가능성을 확인할 수 있었다.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 1991년도 한국자동제어학술회의논문집(국내학술편); KOEX, Seoul; 22-24 Oct. 1991
• /
• pp.1022-1027
• /
• 1991

A new method to control a robot manipulator by neural networks is proposed. The controller is composed of both a PD controller and a neural network-based feedforward controller. MLP(multi-layer perceptron) neural network is used for the feedforward controller and trained by BP(back-propagation) learning rule. Error terms for BP learning rule are composed of the outputs of a PD controller and the acceleration errors of manipulator joints. We compare the proposed method with existing ones and contrast performances of them by simulation. Also, We discuss the real application of the proposed method in consideration of the learning time of the neural network and the time required for sensing the joint acceleration.

• 한국HCI학회:학술대회논문집
• /
• 한국HCI학회 2007년도 학술대회 3부
• /
• pp.322-325
• /
• 2007

Graphics processing unit (GPU) is used for a faster artificial neural network. It is used to implement the matrix multiplication of a neural network to enhance the time performance of a text detection system. Preliminary results produced a 20-fold performance enhancement using an ATI RADEON 9700 PRO board. The parallelism of a GPU is fully utilized by accumulating a lot of input feature vectors and weight vectors, then converting the many inner-product operations into one matrix operation. Further research areas include benchmarking the performance with various hardware and GPU-aware learning algorithms. (c) 2004 Pattern Recognition Society. Published by Elsevier Ltd. All rights reserved.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1991년도 하계학술대회 논문집
• /
• pp.900-904
• /
• 1991

In this paper, our study for the recognition of Hand-written Korean characters, Arabic numbers and alphabets by neural netwoks. This System extracts feature of character by using the MESH feature point of handwritten character, Arabic numbers and alphabets. To reduce the input image data, features are extracted from each input images. A MLP(multi-layer perceptron) with one hidden layer was trained with a modified BEP(back error propagation) algorithm. This method extracts feature sets of the characters directly from the scanner and can enhance computation speed without using the special preprocesses such as size normalization, smoothing, and thinning.

• 한국지능시스템학회논문지
• /
• 제13권1호
• /
• pp.24-30
• /
• 2003

본 논문에서는 혼돈 시스템을 제어하기 위해 웨이블릿 신경 회로망을 예측기로 사용하여 일반형 예측 제어기를 설계하는 방법을 제안한다. 본 논문의 방법에서는 웨이블릿 신경 회로망의 각 파라미터에 대한 학습은 예측 오차를 이용한 경사 하강법에 의해 수행되며, 제어 신호는 웨이블릿 신경 회로망의 출력과 기준 신호 사이의 제어 오차를 최소화함으로써 생성된다. 한편, 모의 실험을 통해 본 논문에서 제안한 제어기를 각각 연속 시간 혼돈 시스템인 Doffing 시스템과 이산 시간 혼돈 시스템인 Henon 시스템에 적용하여 제어기의 효율성을 확인하고 아울러 신경 회로망을 이용한 예측 제어의 결과와 비교함으로써 제어기의 우수성을 검증한다

• 전기전자학회논문지
• /
• 제15권4호
• /
• pp.354-362
• /
• 2011

인공신경망회로에서 아직도 안 풀리는 문제 중 하나는 회로의 처리용량에 관한 것이다. 본 논문은 인공신경망회로의 가장 기본이 되는 하나의 입력과 하나의 출력을 갖은 단층 다단 코어넷을 제안하고 그 처리 용량에 관한 수식을 유도하였다. 제안된 코어넷의 처리 용량으로 p단 입력과 q단 출력을 갖는 코어넷의 처리용량(구현 가능한 함수의 수)은 $a_{p,q}=\frac{1}{2}p(p-1)q^2-\frac{1}{2}(p-2)(3p-1)q+(p-1)(p-2)$ 이며, 입력단 p 값이 짝수이고, 출력단 q가 홀수값이면 추가로 (p-1)(p-2)(q-2)/2 만큼 감해진다. 입력 값으로 3단(level), 출력 값으로 6단을 갖는 1(3)-1(6) 모델을 시뮬레이션하여 분석한 결과, 총 216가지의 함수 조합에서 입력 레벨링 방법으로 cot(x)를 이용하여 82가지의 함수가 구현가능 함을 보였다. 이 모델의 시뮬레이션 결과 80개의 함수가 수렴(구현 가능)하였고, 나머지 수렴되지 않은 함수 중에서 2개의 함수는 무게값 공간에서 무게값 좌표를 미리 계산하여 구현 가능함으로 나와, 총 82개의 구현 가능한 함수가 있음을 보였으며, 이는 위 코어넷 처리용량에 의한 계산 값과 일치하였다.