• 제목/요약/키워드: Layer-by-layer learning

검색결과 642건 처리시간 0.038초

협업 계층을 적용한 합성곱 신경망 기반의 이미지 라벨 예측 알고리즘 (Image Label Prediction Algorithm based on Convolution Neural Network with Collaborative Layer)

  • 이현호;이원진
    • 한국멀티미디어학회논문지
    • /
    • 제23권6호
    • /
    • pp.756-764
    • /
    • 2020
  • A typical algorithm used for image analysis is the Convolutional Neural Network(CNN). R-CNN, Fast R-CNN, Faster R-CNN, etc. have been studied to improve the performance of the CNN, but they essentially require large amounts of data and high algorithmic complexity., making them inappropriate for small and medium-sized services. Therefore, in this paper, the image label prediction algorithm based on CNN with collaborative layer with low complexity, high accuracy, and small amount of data was proposed. The proposed algorithm was designed to replace the part of the neural network that is performed to predict the final label in the existing deep learning algorithm by implementing collaborative filtering as a layer. It is expected that the proposed algorithm can contribute greatly to small and medium-sized content services that is unsuitable to apply the existing deep learning algorithm with high complexity and high server cost.

활성화 함수에 따른 유출량 산정 인공신경망 모형의 성능 비교 (Comparison of Artificial Neural Network Model Capability for Runoff Estimation about Activation Functions)

  • 김마가;최진용;방재홍;윤푸른;김귀훈
    • 한국농공학회논문집
    • /
    • 제63권1호
    • /
    • pp.103-116
    • /
    • 2021
  • Analysis of runoff is substantial for effective water management in the watershed. Runoff occurs by reaction of a watershed to the rainfall and has non-linearity and uncertainty due to the complex relation of weather and watershed factors. ANN (Artificial Neural Network), which learns from the data, is one of the machine learning technique known as a proper model to interpret non-linear data. The performance of ANN is affected by the ANN's structure, the number of hidden layer nodes, learning rate, and activation function. Especially, the activation function has a role to deliver the information entered and decides the way of making output. Therefore, It is important to apply appropriate activation functions according to the problem to solve. In this paper, ANN models were constructed to estimate runoff with different activation functions and each model was compared and evaluated. Sigmoid, Hyperbolic tangent, ReLU (Rectified Linear Unit), ELU (Exponential Linear Unit) functions were applied to the hidden layer, and Identity, ReLU, Softplus functions applied to the output layer. The statistical parameters including coefficient of determination, NSE (Nash and Sutcliffe Efficiency), NSEln (modified NSE), and PBIAS (Percent BIAS) were utilized to evaluate the ANN models. From the result, applications of Hyperbolic tangent function and ELU function to the hidden layer and Identity function to the output layer show competent performance rather than other functions which demonstrated the function selection in the ANN structure can affect the performance of ANN.

에지 클라우드 환경에서 사물인터넷 트래픽 침입 탐지 (Intrusion Detection for IoT Traffic in Edge Cloud)

  • Shin, Kwang-Seong;Youm, Sungkwan
    • 한국정보통신학회논문지
    • /
    • 제24권1호
    • /
    • pp.138-140
    • /
    • 2020
  • As the IoT is applied to home and industrial networks, data generated by the IoT is being processed at the cloud edge. Intrusion detection function is very important because it can be operated by invading IoT devices through the cloud edge. Data delivered to the edge network in the cloud environment is traffic at the application layer. In order to determine the intrusion of the packet transmitted to the IoT, the intrusion should be detected at the application layer. This paper proposes the intrusion detection function at the application layer excluding normal traffic from IoT intrusion detection function. As the proposed method, we obtained the intrusion detection result by decision tree method and explained the detection result for each feature.

미래 네트워크 제공을 위한 기계 학습 기반 스마트 서비스 추상화 계층 설계 (Design of Machine Learning based Smart Service Abstraction Layer for Future Network Provisioning)

  • ;;김경백;최덕재
    • 한국정보처리학회:학술대회논문집
    • /
    • 한국정보처리학회 2016년도 추계학술발표대회
    • /
    • pp.114-116
    • /
    • 2016
  • Recently, SDN and NFV technology have been developed actively and provide enormous flexibility of network provisioning. The future network services would generally involve many different types of services such as hologram games, social network live streaming videos and cloud-computing services, which have dynamic service requirements. To provision networks for future services dynamically and efficiently, SDN/NFV orchestrators must clearly understand the service requirements. Currently, network provisioning relies heavily on QoS parameters such as bandwidth, delay, jitter and throughput, and those parameters are necessary to describe the network requirements of a service. However it is often difficult for users to understand and use them proficiently. Therefore, in order to maintain interoperability and homogeneity, it is required to have a service abstraction layer between users and orchestrators. The service abstraction layer analyzes ambiguous user's requirements for the desired services, and this layer generates corresponding refined services requirements. In this paper, we present our initial effort to design a Smart Service Abstraction Layer (SmSAL) for future network architecture, which takes advantage of machine learning method to analyze ambiguous and abstracted user-friendly input parameters and generate corresponding network parameters of the desired service for better network provisioning. As an initial proof-of-concept implementation for providing viability of the proposed idea, we implemented SmSAL with a decision tree model created by learning process with previous service requests in order to generate network parameters related to various audio and video services, and showed that the parameters are generated successfully.

합성곱 신경망 기초 실습 사례 개발 (Development of Convolutional Neural Network Basic Practice Cases)

  • 허경
    • 실천공학교육논문지
    • /
    • 제14권2호
    • /
    • pp.279-285
    • /
    • 2022
  • 본 논문에서는 비전공자들을 위한 교양과정으로, 기초 합성곱신경망 과목 커리큘럼을 설계하는데 필수적으로 요구되는 합성곱신경망 기초 실습 사례를 개발하였다. 개발된 실습 사례는 합성곱신경망의 동작원리를 이해시키는 데 초점을 두고, 시각화된 전체 과정을 확인할 수 있도록 스프레드시트를 사용하였다. 개발된 실습 사례는 지도학습 방식의 이미지 훈련데이터 생성, 입력층, 컨볼루션층(합성곱층), 풀링층 그리고 출력층을 차례대로 구현하고, 신규 데이터에 대해 합성곱신경망의 성능을 테스트하는 것으로 구성되었다. 본 논문에서 개발한 실습사례를 확장하여 인식하려는 이미지 개수를 확장하거나, 고화질의 이미지에 대한 압축률을 높이는 합성곱신경망을 만드는 기초 실습 사례를 만들 수 있다. 따라서, 본 합성곱신경망 기초 실습 사례의 활용도가 높다고 할 수 있다.

효과적인 패턴 인식을 위한 개선된 Counterpropagation 알고리즘 (An Enhanced Counterpropagation Algorithm for Effective Pattern Recognition)

  • 김광백
    • 한국정보통신학회논문지
    • /
    • 제12권9호
    • /
    • pp.1682-1688
    • /
    • 2008
  • CP(Counterpropagation) 알고리즘은 Kohonen의 경쟁 네트워크와 Grossberg의 아웃스타(Outstar) 구조의 결합으로 이루어진 것으로 패턴 매칭, 패턴 분류, 통계적인 분석 및 데이터 압축 등 활용분야가 다양하고, 다른 신경망 모델에 비해 학습이 매우 빠르다는 장점이 있다. 그러나 CP 알고리즘은 충분한 경쟁층의 수가 설정되지 않아 경쟁층에서 학습이 불안정하고, 다양한 패턴으로 구성된 경우에는 패턴들을 정확히 분류할 수 없는 경우가 발생한다. 그리고 CP 알고리즘은 출력층에서 연결 강도를 조정할 때, 학습률에 따라 학습 및 인식 성능이 좌우된다. 본 논문에서는 효과적인 패턴인식을 위해 다수 경쟁층을 설정하고, 입력 벡터와 승자 뉴런의 대표 벡터간의 차이와 승자 뉴런의 빈도수를 학습률 조정에 반영하고 학습률을 동적으로 조정하여 경쟁층에서 안정적으로 학습되도록 하고, 출력층의 연결강도를 조정할 때 모멘텀(Momentum) 방법을 적용한다. 제안된 CP 학습 성능을 확인하기 위해서 실제 여권에서 추출된 개별 코드를 대상으로 실험한 결과, 개선된 CP 알고리즘이 기존의 CP 알고리즘보다 학습 성능, 분류의 정확성 및 인식 성능이 개선된 것을 확인하였다.

지진 이벤트 분류를 위한 정규화 기법 분석 (Analysis of normalization effect for earthquake events classification)

  • 장수;구본화;고한석
    • 한국음향학회지
    • /
    • 제40권2호
    • /
    • pp.130-138
    • /
    • 2021
  • 본 논문에서는 지진 이벤트 분류를 위한 다양한 정규화 기법 분석 및 효과적인 합성곱 신경망(Convolutional Neural Network, CNN)기반의 네트워크 구조를 제안하였다. 정규화 기법은 신경망의 학습 속도를 개선할 뿐만 아니라 잡음에 강인한 모습을 보여 준다. 본 논문에서는 지진 이벤트 분류를 위한 딥러닝 모델에서 입력 정규화 및 은닉 레이어 정규화가 모델에 미치는 영향을 분석하였다. 또한, 적용 은닉 레이어의 구조에 따른 다양한 실험을 통해 효과적인 모델을 도출하였다. 다양한 모의실험 결과 입력 데이터 정규화 및 제1 은닉 레이어에 가중치 정규화를 적용한 모델이 가장 안정적인 성능 향상을 보여 주었다.

Residual Learning Based CNN for Gesture Recognition in Robot Interaction

  • Han, Hua
    • Journal of Information Processing Systems
    • /
    • 제17권2호
    • /
    • pp.385-398
    • /
    • 2021
  • The complexity of deep learning models affects the real-time performance of gesture recognition, thereby limiting the application of gesture recognition algorithms in actual scenarios. Hence, a residual learning neural network based on a deep convolutional neural network is proposed. First, small convolution kernels are used to extract the local details of gesture images. Subsequently, a shallow residual structure is built to share weights, thereby avoiding gradient disappearance or gradient explosion as the network layer deepens; consequently, the difficulty of model optimisation is simplified. Additional convolutional neural networks are used to accelerate the refinement of deep abstract features based on the spatial importance of the gesture feature distribution. Finally, a fully connected cascade softmax classifier is used to complete the gesture recognition. Compared with the dense connection multiplexing feature information network, the proposed algorithm is optimised in feature multiplexing to avoid performance fluctuations caused by feature redundancy. Experimental results from the ISOGD gesture dataset and Gesture dataset prove that the proposed algorithm affords a fast convergence speed and high accuracy.

고차 뉴런을 이용한 교사 학습기의 Kohonen Feature Map (Using Higher Order Neuron on the Supervised Learning Machine of Kohonen Feature Map)

  • 정종수;하기와라 마사후미
    • 대한전기학회논문지:시스템및제어부문D
    • /
    • 제52권5호
    • /
    • pp.277-282
    • /
    • 2003
  • In this paper we propose Using Higher Order Neuron on the Supervised Learning Machine of the Kohonen Feature Map. The architecture of proposed model adopts the higher order neuron in the input layer of Kohonen Feature Map as a Supervised Learning Machine. It is able to estimate boundary on input pattern space because or the higher order neuron. However, it suffers from a problem that the number of neuron weight increases because of the higher order neuron in the input layer. In this time, we solved this problem by placing the second order neuron among the higher order neuron. The feature of the higher order neuron can be mapped similar inputs on the Kohonen Feature Map. It also is the network with topological mapping. We have simulated the proposed model in respect of the recognition rate by XOR problem, discrimination of 20 alphabet patterns, Mirror Symmetry problem, and numerical letters Pattern Problem.

DenseNet을 활용한 식물 잎 분류 방안 연구 (Classification Method of Plant Leaf using DenseNet)

  • 박용민;강수명;채지훈;이준재
    • 한국멀티미디어학회논문지
    • /
    • 제21권5호
    • /
    • pp.571-582
    • /
    • 2018
  • Recently, development of deep learning has shown better image classification result than human. According to recent research, a hidden layer of deep learning is deeper, and a preservation of extracted features shows good results. However, in the case of general images, the extracted features are clear and easy to sort. This study aims to classify plant leaf images. This plant leaf image has high similarity in each image. Since plant leaf images have high similarity not only between images of different species but also within the same species, classification accuracy is not increased by simply extending the hidden layer or connecting the layers. Therefore, in this paper, we tried to improve the hidden layer of the algorithm called DenseNet which shows the recent excellent classification results, and compare the results of several different modified layers. The proposed method makes it possible to classify plant leaf images collected in a natural environment more easily and accurately than conventional methods. This results in good classification of plant leaf image data including unnecessary noise obtained in a natural environment.