• The Journal of the Acoustical Society of Korea
• /
• v.13 no.1E
• /
• pp.25-36
• /
• 1994

A problem of making a neural network learning self-adaptive to the training set supplied is addressed in this paper. This arises from the aspect in choice of an adequate stepsize for the update of the current weigh vectors according to the training pairs. Related issues in this attempt are raised and fundamentals in neural network learning are introduced. In comparison to the most popular back-propagation scheme, the usefulness and superiority of the proposed weight update algorithm are illustrated by examing the identification of unknown nonlinear systems only from measurements.

• Journal of Korean Society of Industrial and Systems Engineering
• /
• v.44 no.4
• /
• pp.169-176
• /
• 2021

Maritime monitoring requirements have been beyond human operators capabilities due to the broadness of the coverage area and the variety of monitoring activities, e.g. illegal migration, or security threats by foreign warships. Abnormal vessel movement can be defined as an unreasonable movement deviation from the usual trajectory, speed, or other traffic parameters. Detection of the abnormal vessel movement requires the operators not only to pay short-term attention but also to have long-term trajectory trace ability. Recent advances in deep learning have shown the potential of deep learning techniques to discover hidden and more complex relations that often lie in low dimensional latent spaces. In this paper, we propose a deep autoencoder-based clustering model for automatic detection of vessel movement anomaly to assist monitoring operators to take actions on the vessel for more investigation. We first generate gridded trajectory images by mapping the raw vessel trajectories into two dimensional matrix. Based on the gridded image input, we test the proposed model along with the other deep autoencoder-based models for the abnormal trajectory data generated through rotation and speed variation from normal trajectories. We show that the proposed model improves detection accuracy for the generated abnormal trajectories compared to the other models.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.2
• /
• pp.438-455
• /
• 2024

Recently, many unsupervised learning-based models have emerged for Just Noticeable Difference (JND) estimation, demonstrating remarkable improvements in accuracy. However, these models suffer from a significant drawback is that their heavy reliance on handcrafted priors for guidance. This restricts the information for estimating JND simply extracted from regions that are highly related to handcrafted priors, while information from the rest of the regions is disregarded, thus limiting the accuracy of JND estimation. To address such issue, on the one hand, we extract the information for estimating JND in an Adversarial Complementary Learning (ACoL) way and propose an ACoL-JND network to estimate the JND by comprehensively considering the handcrafted priors-related regions and non-related regions. On the other hand, to make the handcrafted priors richer, we take two additional priors that are highly related to JND modeling into account, i.e., Patterned Masking (PM) and Contrast Masking (CM). Experimental results demonstrate that our proposed model outperforms the existing JND models and achieves state-of-the-art performance in both subjective viewing tests and objective metrics assessments.

• Structural Engineering and Mechanics
• /
• v.77 no.1
• /
• pp.47-56
• /
• 2021

This paper deals with damage detection using a Gapped Smoothing Method (GSM) combined with deep learning. Convolutional Neural Network (CNN) is a model of deep learning. CNN has an input layer, an output layer, and a number of hidden layers that consist of convolutional layers. The input layer is a tensor with shape (number of images) × (image width) × (image height) × (image depth). An activation function is applied each time to this tensor passing through a hidden layer and the last layer is the fully connected layer. After the fully connected layer, the output layer, which is the final layer, is predicted by CNN. In this paper, a complete machine learning system is introduced. The training data was taken from a Finite Element (FE) model. The input images are the contour plots of curvature gapped smooth damage index. A free-free beam is used as a case study. In the first step, the FE model of the beam was used to generate data. The collected data were then divided into two parts, i.e. 70% for training and 30% for validation. In the second step, the proposed CNN was trained using training data and then validated using available data. Furthermore, a vibration experiment on steel damaged beam in free-free support condition was carried out in the laboratory to test the method. A total number of 15 accelerometers were set up to measure the mode shapes and calculate the curvature gapped smooth of the damaged beam. Two scenarios were introduced with different severities of the damage. The results showed that the trained CNN was successful in detecting the location as well as the severity of the damage in the experimental damaged beam.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.18 no.3
• /
• pp.360-366
• /
• 2008

The initial fuzzy partitions in fuzzy rule-based classification systems are determined by considering the domain region of each attribute with the given data, and the optimal classification boundaries within the fuzzy partitions can be discovered by tuning their parameters using various learning processes such as neural network, genetic algorithm, and so on. In this paper, we propose a selection method for fuzzy partition based on statistical information to maximize the performance of pattern classification without learning processes where statistical information is used to extract the uncertainty regions (i.e., the regions which the classification boundaries in pattern classification problems are determined) in each input attribute from the numerical data. Moreover the methods for extracting the candidate rules which are associated with the partition intervals generated by statistical information and for minimizing the coupling problem between the candidate rules are additionally discussed. In order to show the effectiveness of the proposed method, we compared the classification accuracy of the proposed with those of conventional methods on the IRIS and New Thyroid Cancer data. From experimental results, we can confirm the fact that the proposed method only considering statistical information of the numerical patterns provides equal to or better classification accuracy than that of the conventional methods.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.814-819
• /
• 2004

Soft engineering, based on symbiotic coexistence of human, machines and environment, is a new engineering field to explore the proper technology and the proper way of engineering. To explore soft engineering intents easily, various robot projects at Pusan National University conducted are presented. Thought experiment, interactive e-leaning, rapid prototyping engineering, biomimciry, tangibility, and ubiquity are concepts to be explored. Thought experiments projects are organized and performed, which include robot assembly game, Turing test, and robotics in science fiction. "Junk robot project" and "ubiquitous Pusan National University (u-PNU) project" have been organized. Also, bug robot project, interactive robot project, and interactive emotional robot projects are introduced. Weekly science fiction films are shown and discussed.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.23 no.4
• /
• pp.141-149
• /
• 1998

One of the difficulties in using the current Boolean-based information retrieval systems is that it is hard for a user, especially a novice, to formulate an effective Boolean query. One solution to this problem is to let the system formulate a query for a user from his relevance feedback documents in this research, an intelligent query reformulation mechanism based on ID3 is proposed and the sensitivity of its retrieval effectiveness, i.e., recall, precision, and E-measure, to various input settings is analyzed. The parameters in the input settings is the number of relevant documents. Experiments conducted on the test set of Medlars revealed that the effectiveness of the proposed system is in fact sensitive to the number of the initial relevant documents. The case with two or more initial relevant documents outperformed the case with one initial relevant document with statistical significances. It is our conclusion that formulation of an effective query in the proposed system requires at least two relevant documents in its initial input set.

• Journal of Korea Water Resources Association
• /
• v.53 no.6
• /
• pp.395-408
• /
• 2020

This paper aims to develop and apply three different machine learning approaches (i.e., artificial neural networks (ANN), adaptive neuro-fuzzy inference systems (ANFIS), and wavelet-based neural networks (WNN)) combined with an evolutionary optimization algorithm and the k-fold cross validation for multi-step (days) streamflow forecasting at the catchment located in Algeria, North Africa. The ANN and ANFIS models yielded similar performances, based on four different statistical indices (i.e., root mean squared error (RMSE), Nash-Sutcliffe efficiency (NSE), correlation coefficient (R), and peak flow criteria (PFC)) for training and testing phases. The values of RMSE and PFC for the WNN model (e.g., RMSE = 8.590 ㎥/sec, PFC = 0.252 for (t+1) day, testing phase) were lower than those of ANN (e.g., RMSE = 19.120 ㎥/sec, PFC = 0.446 for (t+1) day, testing phase) and ANFIS (e.g., RMSE = 18.520 ㎥/sec, PFC = 0.444 for (t+1) day, testing phase) models, while the values of NSE and R for WNN model were higher than those of ANNs and ANFIS models. Therefore, the new approach can be a robust tool for multi-step (days) streamflow forecasting in the Seybous River, Algeria.

• The Journal of Society for e-Business Studies
• /
• v.25 no.1
• /
• pp.1-12
• /
• 2020

Recently, research using machine learning have shown remarkable results in various domains, leading to the fact that leaning-based dispatchers have intrigued interest in both academia and industry. To improve the performance of the dispatcher, each dispatch decision needs to be evaluated in detail. However, existing studies on visualization techniques for manufacturing lines have mainly focused on illustrating the performance indicators or abnormal patterns. In this paper, we propose a monitoring system that displays a variety of information about the manufacturing line along with alternatives at the time of each dispatching decision being made. Furthermore, the proposed system effectively represents the cause of the idle time of resources and the change of the performance index over time.

• Journal of Information Processing Systems
• /
• v.16 no.1
• /
• pp.6-29
• /
• 2020

Biometrics identification using multiple modalities has attracted the attention of many researchers as it produces more robust and trustworthy results than single modality biometrics. In this paper, we present a novel multimodal recognition system that trains a deep learning network to automatically learn features after extracting multiple biometric modalities from a single data source, i.e., facial video clips. Utilizing different modalities, i.e., left ear, left profile face, frontal face, right profile face, and right ear, present in the facial video clips, we train supervised denoising auto-encoders to automatically extract robust and non-redundant features. The automatically learned features are then used to train modality specific sparse classifiers to perform the multimodal recognition. Moreover, the proposed technique has proven robust when some of the above modalities were missing during the testing. The proposed system has three main components that are responsible for detection, which consists of modality specific detectors to automatically detect images of different modalities present in facial video clips; feature selection, which uses supervised denoising sparse auto-encoders network to capture discriminative representations that are robust to the illumination and pose variations; and classification, which consists of a set of modality specific sparse representation classifiers for unimodal recognition, followed by score level fusion of the recognition results of the available modalities. Experiments conducted on the constrained facial video dataset (WVU) and the unconstrained facial video dataset (HONDA/UCSD), resulted in a 99.17% and 97.14% Rank-1 recognition rates, respectively. The multimodal recognition accuracy demonstrates the superiority and robustness of the proposed approach irrespective of the illumination, non-planar movement, and pose variations present in the video clips even in the situation of missing modalities.