• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Journal of Korea Society of Digital Industry and Information Management
• /
• v.19 no.4
• /
• pp.127-136
• /
• 2023

Many research results on the traffic enforcement of illegal driving of twowheeled motor vehicles using license plate recognition are introduced. Deep learning convolutional neural networks can be used for character and word recognition of license plates because of better generalization capability compared to traditional Backpropagation neural networks. In the plates of twowheeled motor vehicles, the interdependent government and city words are included. If we implement the mutually independent word recognizers using error correction rules for two word recognition results, efficient license plate recognition results can be derived. The CPU based convolutional neural network without library under real time processing has an advantage of low cost real application compared to GPU based convolutional neural network with library. In this paper twowheeled motor vehicle license plate recognition algorithm is introduced using CPU based deep-learning convolutional neural network. The experimental results show that the proposed plate recognizer has 96.2% success rate for outdoor twowheeled motor vehicle images in real time.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.11
• /
• pp.5427-5445
• /
• 2019

While deep neural networks have achieved remarkable performance in representation learning, a huge amount of labeled training data are usually required by supervised deep models such as convolutional neural networks. In this paper, we propose a new representation learning method, namely generative adversarial networks (GAN) based bagging deep convolutional autoencoders (GAN-BDCAE), which can map data to diverse hierarchical representations in an unsupervised fashion. To boost the size of training data, to train deep model and to aggregate diverse learning machines are the three principal avenues towards increasing the capabilities of representation learning of neural networks. We focus on combining those three techniques. To this aim, we adopt GAN for realistic unlabeled sample generation and bagging deep convolutional autoencoders (BDCAE) for robust feature learning. The proposed method improves the discriminative ability of learned feature embedding for solving subsequent pattern recognition problems. We evaluate our approach on three standard benchmarks and demonstrate the superiority of the proposed method compared to traditional unsupervised learning methods.

• Journal of Semiconductor Engineering
• /
• v.1 no.1
• /
• pp.31-37
• /
• 2020

We present an approximate DRAM architecture for energy-efficient deep learning. Our key premise is that by bounding memory errors to non-critical information, we can significantly reduce DRAM refresh energy without compromising recognition accuracy of deep neural networks. To validate the key premise, we make extensive Monte-Carlo simulations for several well-known convolutional neural networks such as LeNet, ConvNet and AlexNet with the input of MINIST, CIFAR-10, and ImageNet, respectively. We assume that the highest-order 8-bits (in single precision) and 4-bits (in half precision) are protected from retention errors under the proposed architecture and then, randomly inject bit-errors to unprotected bits with various bit-error-rates. Here, recognition accuracies of the above convolutional neural networks are successfully maintained up to the 10^-5-order bit-error-rate. We simulate DRAM energy during inference of the above convolutional neural networks, where the proposed architecture shows the possibility of considerable energy saving up to 10 ~ 37.5% of total DRAM energy.

• Journal of the Korean Society of Industry Convergence
• /
• v.23 no.3
• /
• pp.515-521
• /
• 2020

Deep learning methods have been effectively used to provide great improvement in various research fields such as machine learning, image processing and computer vision. One of the most frequently used deep learning methods in image processing is the convolutional neural networks. Compared to the traditional artificial neural networks, convolutional neural networks do not use the predefined kernels, but instead they learn data specific kernels. This property makes them to be used as feature extractors as well. In this study, we compared the quality of CNN features for traditional texture feature extraction methods. Experimental results demonstrate the superiority of the CNN features. Additionally, the recognition process and result of a pioneering CNN on MNIST database are presented.

• Journal of Information Technology Services
• /
• v.16 no.4
• /
• pp.149-160
• /
• 2017

Sentiment analysis is the process of determining whether a piece of document, text or conversation is positive, negative, neural or other emotion. Sentiment analysis has been applied for several real-world applications, such as chatbot. In the last five years, the practical use of the chatbot has been prevailing in many field of industry. In the chatbot applications, to recognize the user emotion, sentiment analysis must be performed in advance in order to understand the intent of speakers. The specific emotion is more than describing positive or negative sentences. In light of this context, we propose deep learning models for conducting multi-class sentiment analysis for identifying speaker's emotion which is categorized to be joy, fear, guilt, sad, shame, disgust, and anger. Thus, we develop convolutional neural network (CNN), long short term memory (LSTM), and multi-layer neural network models, as deep neural networks models, for detecting emotion in a sentence. In addition, word embedding process was also applied in our research. In our experiments, we have found that long short term memory (LSTM) model performs best compared to convolutional neural networks and multi-layer neural networks. Moreover, we also show the practical applicability of the deep learning models to the sentiment analysis for chatbot.

• Journal of Information Technology Services
• /
• v.15 no.4
• /
• pp.85-95
• /
• 2016

Recently, deep learning and machine learning have attracted considerable attention and many supporting frameworks appeared. In artificial intelligence field, a large body of research is underway to apply the relevant knowledge for complex problem-solving, necessitating the application of various learning algorithms and training methods to artificial intelligence systems. In addition, there is a dearth of performance evaluation of decision making agents. The decision making agent that can find optimal solutions by using reinforcement learning methods designed through this research can collect raw pixel data observed from dynamic environments and make decisions by itself based on the data. The decision making agent uses convolutional neural networks to classify situations it confronts, and the data observed from the environment undergoes preprocessing before being used. This research represents how the convolutional neural networks and the decision making agent are configured, analyzes learning performance through a value-based algorithm and a policy-based algorithm : a Deep Q-Networks and a Policy Gradient, sets forth their differences and demonstrates how the convolutional neural networks affect entire learning performance when using pixel data. This research is expected to contribute to the improvement of artificial intelligence systems which can efficiently find optimal solutions by using features extracted from raw pixel data.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.5
• /
• pp.728-733
• /
• 2018

Deep learning has been actively studied for predicting protein secondary structure based only on the sequence information of the amino acids constituting the protein. In this paper, we compared the performances of the convolutional neural networks of various structures to predict the protein secondary structure. To investigate the optimal depth of the layer of neural network for the prediction of protein secondary structure, the performance according to the number of layers was investigated. We also applied the structure of GoogLeNet and ResNet which constitute building blocks of many image classification methods. These methods extract various features from input data, and smooth the gradient transmission in the learning process even using the deep layer. These architectures of convolutional neural networks were modified to suit the characteristics of protein data to improve performance.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.3
• /
• pp.1228-1248
• /
• 2020

Steganalysis & steganography have witnessed immense progress over the past few years by the advancement of deep convolutional neural networks (DCNN). In this paper, we analyzed current research states from the latest image steganography and steganalysis frameworks based on deep learning. Our objective is to provide for future researchers the work being done on deep learning-based image steganography & steganalysis and highlights the strengths and weakness of existing up-to-date techniques. The result of this study opens new approaches for upcoming research and may serve as source of hypothesis for further significant research on deep learning-based image steganography and steganalysis. Finally, technical challenges of current methods and several promising directions on deep learning steganography and steganalysis are suggested to illustrate how these challenges can be transferred into prolific future research avenues.

• Journal of Korea Multimedia Society
• /
• v.21 no.11
• /
• pp.1244-1253
• /
• 2018

The color of traffic light is sensitive to various illumination conditions. Especially it loses the hue information when oversaturation happens on the lighting area. This paper proposes a traffic light recognition method robust to these illumination variations. The method consists of two steps of traffic light detection and recognition. It just uses the intensity and saturation in the first step of traffic light detection. It delays the use of hue information until it reaches to the second step of recognizing the signal of traffic light. We utilized a deep learning technique in the second step. We designed a deep convolutional neural network(DCNN) which is composed of three convolutional networks and two fully connected networks. 12 video clips were used to evaluate the performance of the proposed method. Experimental results show the performance of traffic light detection reporting the precision of 93.9%, the recall of 91.6%, and the recognition accuracy of 89.4%. Considering that the maximum distance between the camera and traffic lights is 70m, the results shows that the proposed method is effective.