• Journal of Electrical Engineering and Technology
• /
• v.10 no.2
• /
• pp.699-707
• /
• 2015

Although digit character recognition has got a significant improvement in recent years, it is still challenging to achieve satisfied result if the data contains an amount of distracting factors. This paper proposes a novel digit character recognition approach using a multi-layer hierarchical model, Hybrid Restricted Boltzmann Machines (HRBMs), which allows the learning architecture to be robust to background distracting factors. The insight behind the proposed model is that useful high-level features appear more frequently than distracting factors during learning, thus the high-level features can be decompose into hybrid hierarchical structures by using only small label information. In order to extract robust and compact features, a stochastic 0-1 layer is employed, which enables the model's hidden nodes to independently capture the useful character features during training. Experiments on the variations of Mixed National Institute of Standards and Technology (MNIST) dataset show that improvements of the multi-layer hierarchical model can be achieved by the proposed method. Finally, the paper shows the proposed technique which is used in a real-world application, where it is able to identify digit characters under various complex background images.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.10
• /
• pp.497-504
• /
• 2024

Generative Adversarial Networks (GANs) are typically described as a two-player game between a generator and a discriminator, where the generator aims to produce realistic data, and the discriminator tries to distinguish between real and generated data. However, this setup often leads to mode collapse, where the generator produces limited variations in the data, failing to capture the full range of the target data distribution. This paper proposes a new training system to mitigate the mode collapse problem. Specifically, it extends the traditional two-player game of GANs into a multi-player game and introduces a peer-evaluation method to effectively train multiple GANs. In the peer-evaluation process, the generated samples from each GANs are evaluated by the other players. This provides external feedback, serving as an additional standard that helps GANs recognize mode failure. This cooperative yet competitive training method encourages the generators to explore and capture a broader range of the data distribution, mitigating mode collapse problem. This paper explains the detailed algorithm for peer-evaluation based multi-GANs training and validates the performance through experiments.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.21 no.5
• /
• pp.274-286
• /
• 2022

Multi-object tracking has been studied for a long time under computer vision and plays a critical role in applications such as autonomous driving and driving assistance. Multi-object tracking techniques generally consist of a detector that detects objects and a tracker that tracks the detected objects. Various publicly available datasets allow us to train a detector model without much effort. However, there are relatively few publicly available datasets for training a tracker model, and configuring own tracker datasets takes a long time compared to configuring detector datasets. Hence, the detector is often developed separately with a tracker module. However, the separated tracker should be adjusted whenever the former detector model is changed. This study proposes a system that can train a model that performs detection and tracking simultaneously using only the detector training datasets. In particular, a Siam network with augmentation is used to compose the detector and tracker. Experiments are conducted on public datasets to verify that the proposed algorithm can formulate a real-time multi-object tracker comparable to the state-of-the-art tracker models.

• Computers and Concrete
• /
• v.30 no.5
• /
• pp.301-310
• /
• 2022

Cement-based sensors have been widely used as structural health monitoring systems, however, their long-term sensing performance have not actively investigated. In this study, a deep learning-based methodology is adopted to predict the long-term piezoresistive properties of cement-based sensors. Samples with different multi-walled carbon nanotube contents (0.1, 0.3, and 0.5 wt.%) are fabricated, and piezoresistive tests are conducted over 10,000 loading cycles to obtain the training data. Time-dependent degradation is predicted using a modified long short-term memory (LSTM) model. The effects of different model variables including the amount of training data, number of epochs, and dropout ratio on the accuracy of predictions are analyzed. Finally, the effectiveness of the proposed approach is evaluated by comparing the predictions for long-term piezoresistive sensing performance with untrained experimental data. A sensitivity of 6% is experimentally examined in the sample containing 0.1 wt.% of MWCNTs, and predictions with accuracy up to 98% are found using the proposed LSTM model. Based on the experimental results, the proposed model is expected to be applied in the structural health monitoring systems to predict their long-term piezoresistice sensing performances during their service life.

• Asia pacific journal of information systems
• /
• v.32 no.2
• /
• pp.226-248
• /
• 2022

The existing model for sentiment analysis of product reviews learned from past data and new data was labeled based on training. But new data was never used by the existing system for making a decision. The proposed Aspect-based multi-agent Deep Reinforcement learning Sentiment Analysis (ADRSA) model learned from its very first data without the help of any training dataset and labeled a sentence with aspect category and sentiment polarity. It keeps on learning from the new data and updates its knowledge for improving its intelligence. The decision of the proposed system changed over time based on the new data. So, the accuracy of the sentiment analysis using deep reinforcement learning was improved over supervised learning and unsupervised learning methods. Hence, the sentiments of premium customers on a particular site can be explored to other customers effectively. A dynamic environment with a strong knowledge base can help the system to remember the sentences and usage State Action Reward State Action (SARSA) algorithm with Bidirectional Encoder Representations from Transformers (BERT) model improved the performance of the proposed system in terms of accuracy when compared to the state of art methods.

• Journal of Digital Convergence
• /
• v.17 no.12
• /
• pp.243-248
• /
• 2019

Multi-Task Learning(MTL) is a training method that trains a single neural network with multiple tasks influences each other. In this paper, we compare performance of MTL Named entity recognition(NER) model trained with Korean traditional culture corpus and other NER model. In training process, each Bi-LSTM layer of Part of speech tagging(POS-tagging) and NER are propagated from a Bi-LSTM layer to obtain the joint loss. As a result, the MTL based Bi-LSTM model shows 1.1%~4.6% performance improvement compared to single Bi-LSTM models.

• Convergence Security Journal
• /
• v.24 no.3
• /
• pp.41-49
• /
• 2024

A tokenizer, which is a key component of the Transformer model, lacks the ability to effectively comprehend numerical data. Therefore, to develop a Transformer-based intrusion detection model that can operate within a real-world network environment by training packet payloads as sentences, it is necessary to convert the hexadecimal packet payloads into a character-based format. In this study, we applied three character encoding methods to convert packet payloads into numeric or character format and analyzed how detection performance changes when training them on transformer architecture. The experimental dataset was generated by extracting packet payloads from PCAP files included in the UNSW-NB15 dataset, and the RoBERTa was used as the training model. The experimental results demonstrate that the ISO-8859-1 encoding scheme achieves the highest performance in both binary and multi-class classification. In addition, when the number of tokens is set to 512 and the maximum number of epochs is set to 15, the multi-class classification accuracy is improved to 88.77%.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.3
• /
• pp.1477-1491
• /
• 2017

Inspired by the multi-scale nature of hippocampal place cells, a biologically inspired model based on a multi-scale spatial representation for goal-directed navigation is proposed in order to achieve robotic spatial cognition and autonomous navigation. First, a map of the place cells is constructed in different scales, which is used for encoding the spatial environment. Then, the firing rate of the place cells in each layer is calculated by the Gaussian function as the input of the Q-learning process. The robot decides on its next direction for movement through several candidate actions according to the rules of action selection. After several training trials, the robot can accumulate experiential knowledge and thus learn an appropriate navigation policy to find its goal. The results in simulation show that, in contrast to the other two methods(G-Q, S-Q), the multi-scale model presented in this paper is not only in line with the multi-scale nature of place cells, but also has a faster learning potential to find the optimized path to the goal. Additionally, this method also has a good ability to complete the goal-directed navigation task in large space and in the environments with obstacles.

• IE interfaces
• /
• v.23 no.4
• /
• pp.311-318
• /
• 2010

In the daily multi-reservoir operating problem, monthly storage targets can be used as principal operational guidelines. In this study, we tested the use of a simple back-propagation Artificial Neural Network (ANN) model to derive monthly storage guideline for daily Coordinated Multi-reservoir Operating Model (CoMOM) of the Han-River basin. This approach is based on the belief that the optimum solution of the daily CoMOM has a good performance, and the ANN model trained with the results of daily CoMOM would produce effective monthly operating guidelines. The optimum results of daily CoMOM is used as the training set for the back-propagation ANN model, which is designed to derive monthly reservoir storage targets in the basin. For the input patterns of the ANN model, we adopted the ratios of initial storage of each dam to the storage of Paldang dam, ratios of monthly expected inflow of each dam to the total inflow of the whole basin, ratios of monthly demand at each dam to the total demand of the whole basin, ratio of total storage of the whole basin to the active storage of Paldang dam, and the ratio of total inflow of the whole basin to the active storage of the whole basin. And the output pattern of ANN model is the optimal final storages that are generated by the daily CoMOM. Then, we analyzed the performance of the ANN model by using a real-time simulation procedure for the multi-reservoir system of the Han-river basin, assuming that historical inflows from October 1st, 2004 to June 30th, 2007 (except July, August, September) were occurred. The simulation results showed that by utilizing the monthly storage target provided by the ANN model, we could reduce the spillages, increase hydropower generation, and secure more water at the end of the planning horizon compared to the historical records.

• Proceedings of the Korea Electromagnetic Engineering Society Conference
• /
• 2003.11a
• /
• pp.594-598
• /
• 2003

In this paper, we analyze the performance of UWB(Ultra-WideBand) communication system employing Bi-phase modulation and RAKE Receiver under the MAI(Multiple Access Interference) and the OSI(Other System Interference) environment. Using the multi-path channel model recommended by IEEE P802.15.TG3a, the performance degradation Is described with the number of users, the number of RAKE fingers and training sequences. To meet BER 10e-4 for 20 users at the same time, the number of RAKE fingers are proposed from 3 to 32. And the number of training sequences are limited less than 8 to keep the channel estimation error within 3dB