• Annual Conference on Human and Language Technology
• /
• 2012.10a
• /
• pp.142-147
• /
• 2012

본 논문에서는 신경망을 재귀적으로 사용하여 문장에서 지시 표현을 추출하고 분석하는 방법에 대해서 제안한다. 임의의 문장이 들어올 때, 문장을 구성하는 각 단어들은 통사론적 자질 벡터와 의미론적 자질 벡터로 나눌 수 있다. 이들 벡터들의 쌍을 인자로써 입력받는 신경망 구조를 제시할 수 있으며, 신경망의 출력 결과는 다시 재귀적으로 쌍인자 신경망에 입력으로써 주입된다. 신경망을 재귀적으로 학습시킴으로써, 문장 내의 지시 표현을 추출할 수 있다. 쌍인자 신경망 파싱 모델의 성능을 측정했고, 제안한 모델의 문제점과 가능성에 대해서 관찰하였다.

• The Journal of the Acoustical Society of Korea
• /
• v.17 no.4
• /
• pp.3-13
• /
• 1998

본 논문은 음성 다이어링 시스템을 구현하기 위한 한국어 단독 숫자음 및 연속 숫 자음 인식에 관한 것이다. 단독 숫자음의 인식은 미지의 입력 음성을 재귀 신경망을 이용하 여 모델링된 각 모델에 인가하고, 신경 회로망의 출력 노드의 상태열을 검사하여 적절한 상 태 전이를 하며 최고의 확률값을 출력하는 모델을 인식된 결과로 출력한다. 연속 숫자음의 인식은 미지의 연속 숫자음을 재귀 신경 회로망을 이용한 연속 숫자음 모델에 입력하고, 신 경 회로망의 출력에 대하여 적절한 상태 전이에 대한 검사와 레벨 빌딩(Level Building)을 수행하여 최소의 오차를 가지는 모델열을 인식된 결과로 출력한다. 재귀 신경 회로망을 이 용하여 음절 모델을 만드는 과정에서 재귀 노드는 예상치가 주어지지 않으므로 신경 회로망 의 학습에서 제외되어 현저한 학습 속도의 저하를 가져온다. 따라서 본 논문에서는 재귀 신 경 회로망의 학습 속도를 향상시키기 위한 2가지 방법을 제안 한다. 첫 번째는 재귀 신경 회로망의 재귀 노드의 예상치를 실험적으로 주어줌으로써 학습 속도의 향상을 도모하였다. 두 번째는 음절 모델의 출력노드의 개수와 음절 모델의 세그먼트 경계를 알고리듬을 이용하 여 자동적으로 조절하였다. 실험결과, 단독어의 경우 음절 '에'에 포함하는 한국어 11개의 숫 자음에 대하여 화자 종속의 경우 97.3%, 화자 독립의 경우 80.5%의 인식률을 얻었으며, 연 속 숫자음의 경우는 21종류의 연속 숫자음에 대하여 화자 종속에서 88.2%, 화자 독립의 경 우 81.3%의 인식률을 얻을 수 있었다.

• Journal of the Korea Society of Computer and Information
• /
• v.17 no.3
• /
• pp.11-25
• /
• 2012

We present the method to infer Context-Free Grammars by applying genetic algorithm to the Binary Third-order Recurrent Neural Networks(BTRNN). BTRNN is a multiple-layered architecture of recurrent neural networks, each of which is corresponding to an input symbol, and is combined with external stack. All parameters of BTRNN are represented as binary numbers and each state transition is performed with any stack operation simultaneously. We apply Genetic Algorithm to BTRNN chromosomes and obtain the optimal BTRNN inferring context-free grammar of positive and negative input patterns. This proposed method infers BTRNN, which includes the number of its states equal to or less than those of existing methods of Discrete Recurrent Neural Networks, with less examples and less learning trials. Also BTRNN is superior to the recent method of chromosomes representing grammars at recognition time complexity because of performing deterministic state transitions and stack operations at parsing process. If the number of non-terminals is p, the number of terminals q, the length of an input string k, and the max number of BTRNN states m, the parallel processing time is O(k) and the sequential processing time is O(km).

• Journal of Intelligence and Information Systems
• /
• v.12 no.1
• /
• pp.107-123
• /
• 2006

We propose the Generalized Binary Second-order Recurrent Neural Networks(GBSRNNf) being equivalent to regular grammars and ?how the implementation of lexical analyzer recognizing the regular languages by using it. All the equivalent representations of regular grammars can be implemented in circuits by using GSBRNN, since it has binary-valued components and shows the structural relationship of a regular grammar. For a regular grammar with the number of symbols m, the number of terminals p, the number of nonterminals q, and the length of input string k, the size of the corresponding GBSRNN is $O(m(p+q)^2)$ and its parallel processing time is O(k) and its sequential processing time, $O(k(p+q)^2)$.

• Journal of Broadcast Engineering
• /
• v.24 no.4
• /
• pp.633-642
• /
• 2019

Recently, the convolution neural network (CNN) model at a single image super-resolution (SISR) have been very successful. The residual learning method can improve training stability and network performance in CNN. In this paper, we propose a SISR using recursive residual network architecture by introducing dense skip connections for learning nonlinear mapping from low-resolution input image to high-resolution target image. The proposed SISR method adopts a method of the recursive residual learning to mitigate the difficulty of the deep network training and remove unnecessary modules for easier to optimize in CNN layers because of the concise and compact recursive network via dense skip connection method. The proposed method not only alleviates the vanishing-gradient problem of a very deep network, but also get the outstanding performance with low complexity of neural network, which allows the neural network to perform training, thereby exhibiting improved performance of SISR method.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.3
• /
• pp.27-38
• /
• 2000

This paper proposes how the recurrent neural network controller for a Khepera mobile robot with an obstacle avoiding ability can be determined by co-adaptation of the evolution and learning, The proposed co-adaptation scheme consists of two folds： a population of NN controllers are evolved by the genetic algorithm so that the degree of obstacle avoidance might be reduced through the global searching and each NN controller is trained by CRBP learning so that the running behavior is adapted to its outer environment through the local searching. Experimental results shows that the NN controller coadapted by evolution and learning outperforms its non-learning equivalent evolved by only genetic algorithm in both the ability of obstacle avoidance and the convergence speed reaching to the required running behavior.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.3
• /
• pp.349-354
• /
• 2020

The preliminary investigation to expand the green space requires a lot of cost and time. In this paper, we solve the problem by measuring the ratio of green space in a specific region through a convolutional neural network based the green space classification using Google Earth images. First, the proposed method collects various region images in Google Earth and learns them by using the convolutional neural network. The proposed method divides the image recursively to measure the green space ratio of the specific region, and it determines whether the divided image is green space using a trained convolutional neural network model, and then the green space ratio is calculated using the regions determined as the green space. Experimental results show that the proposed method shows high performance in measuring green space ratios in various regions.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.21 no.4
• /
• pp.475-480
• /
• 2011

In this paper, we propose a collision risk assesment system. First, using Kalman Filter, we estimate the information of pedestrian, and second, we compute the collision probability using Monte Carlo Simulations(MCS) and neural network(NN). And we update the collision risk using time history which is called belief. Belief update consider not only output of Kalman Filter of only current time step but also output of Kalman Filter up to the first time step to current time step. The computer simulations will be shown the validity of our proposed method.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.378-390
• /
• 2022

With the recent development of deep composite multiplication neural network learning, deep learning techniques applied to single-image super-resolution have shown good results, and the strong expression ability of deep networks has enabled complex nonlinear mapping between low-resolution and high-resolution images. However, there are limitations in applying it to real-time or low-power devices with increasing parameters and computational amounts due to excessive use of composite multiplication neural networks. This paper uses blocks that extract hierarchical characteristics little by little using information distillation and suggests the Recursive Distillation Super Resolution Network (RDSRN), a lightweight network that improves performance by making more accurate high frequency components through high frequency residual purification blocks. It was confirmed that the proposed network restores images of similar quality compared to RDN, restores images 3.5 times faster with about 32 times fewer parameters and about 10 times less computation, and produces 0.16 dB better performance with about 2.2 times less parameters and 1.8 times faster processing time than the existing lightweight network CARN.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.580-583
• /
• 2018

본 논문에서는 SeqGAN 모델을 사용하여 한국어 시를 자동 생성해 보았다. SeqGAN 모델은 문장 생성을 위해 재귀 신경망과 강화 학습 알고리즘의 하나인 정책 그라디언트(Policy Gradient)와 몬테카를로 검색(Monte Carlo Search, MC) 기법을 생성기에 적용하였다. 시 문장을 자동 생성하기 위한 학습 데이터로는 사랑을 주제로 작성된 시를 사용하였다. SeqGAN 모델을 사용하여 자동 생성된 시는 동일한 구절이 여러번 반복되는 문제를 보였지만 한국어 텍스트 생성에 있어 SeqGAN 모델이 적용 가능함을 확인하였다.