• Journal of the Korean Institute of Intelligent Systems
• /
• v.12 no.2
• /
• pp.129-134
• /
• 2002

Operating the nuclear power generations safely is not easy way because nuclear power generations are very complicated systems. In the main control room of the nuclear power generations, about 4000 numbers of alarms and monitoring devices are equipped to handle the signals corresponding to operating equipments. Thus, operators have to deal with massive information and to analyze the situation immediately. If they could not achieve these task, then they should make big problem in the power generations. Owing to too many variables, operators could be also in the uncontrolled situation. So in this paper, the fault diagnosis system is designed using 2-steps neural networks. This diagnosis method is based on the pattern of the principal variables which could represent the type and severity of faults.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.2
• /
• pp.729-748
• /
• 2021

Vocal detection is one of the fundamental steps in musical information retrieval. Typically, the detection process consists of feature extraction and classification steps. Recently, neural networks are shown to outperform traditional classifiers. In this paper, we report our study on how to improve detection accuracy further by carefully choosing the parameters of the deep network model. Through experiments, we conclude that a feature-classifier model is still better than an end-to-end model. The recommended model uses a spectrogram as the input plane and the classifier is an 18-layer convolutional neural network (CNN). With this arrangement, when compared with existing literature, the proposed model improves the accuracy from 91.8% to 94.1% in Jamendo dataset. As the dataset has an accuracy of more than 90%, the improvement of 2.3% is difficult and valuable. If even higher accuracy is required, the ensemble learning may be used. The recommend setting is a majority vote with seven proposed models. Doing so, the accuracy increases by about 1.1% in Jamendo dataset.

• Journal of Broadcast Engineering
• /
• v.8 no.2
• /
• pp.172-180
• /
• 2003

In this paper, we propose fast and precise extracting & tracking algorithm based on general camera and frame grabber for facial motion capture animation. Proposed algorithm consists of two steps. extracting and tracking. The former is to separate multiple markers from input image using region merging based on neural networks. The latter Is to track extracted multiple markers at each frame using tracking algorithm based on neural networks. In the experiment, we could remove noise and reduce processing time in the step of extraction. In addition, we could have good tracking results in the low frame rates.

• Geophysics and Geophysical Exploration
• /
• v.26 no.3
• /
• pp.114-125
• /
• 2023

The physics-informed neural network (PINN) has been proposed to overcome the limitations of various numerical methods used to solve partial differential equations (PDEs) and the drawbacks of purely data-driven machine learning. The PINN directly applies PDEs to the construction of the loss function, introducing physical constraints to machine learning training. This technique can also be applied to wave equation modeling. However, to solve the wave equation using the PINN, second-order differentiations with respect to input data must be performed during neural network training, and the resulting wavefields contain complex dynamical phenomena, requiring careful strategies. This tutorial elucidates the fundamental concepts of the PINN and discusses considerations for wave equation modeling using the PINN approach. These considerations include spatial coordinate normalization, the selection of activation functions, and strategies for incorporating physics loss. Our experimental results demonstrated that normalizing the spatial coordinates of the training data leads to a more accurate reflection of initial conditions in neural network training for wave equation modeling. Furthermore, the characteristics of various functions were compared to select an appropriate activation function for wavefield prediction using neural networks. These comparisons focused on their differentiation with respect to input data and their convergence properties. Finally, the results of two scenarios for incorporating physics loss into the loss function during neural network training were compared. Through numerical experiments, a curriculum-based learning strategy, applying physics loss after the initial training steps, was more effective than utilizing physics loss from the early training steps. In addition, the effectiveness of the PINN technique was confirmed by comparing these results with those of training without any use of physics loss.

• International Journal of Internet, Broadcasting and Communication
• /
• v.12 no.2
• /
• pp.45-50
• /
• 2020

A Language model with neural networks commonly trained with likelihood loss. Such that the model can learn the sequence of human text. State-of-the-art results achieved in various language generation tasks, e.g., text summarization, dialogue response generation, and text generation, by utilizing the language model's next token output probabilities. Monotonous and boring outputs are a well-known problem of this model, yet only a few solutions proposed to address this problem. Several decoding techniques proposed to suppress repetitive tokens. Unlikelihood training approached this problem by penalizing candidate tokens probabilities if the tokens already seen in previous steps. While the method successfully showed a less repetitive generated token, the method has a large memory consumption because of the training need a big vocabulary size. We effectively reduced memory footprint by encoding words as sequences of subword units. Finally, we report competitive results with token level unlikelihood training in several automatic evaluations compared to the previous work.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.11
• /
• pp.2075-2082
• /
• 2006

This paper presents modeling and optimization techniques for hish efficiency solar cell process on single-crystalline float zone (FZ) wafers. Among a sequence of multiple steps of fabrication, the followings are the most sensitive steps for the contact formation: 1) Emitter formation by diffusion; 2) Anti-reflection-coating (ARC) with silicon nitride using plasma-enhanced chemical vapor deposition (PECVD); 3) Screen-printing for front and back metalization; and 4) Contact formation by firing. In order to increase the performance of solar cells in terms of efficiency, the contact formation process is modeled and optimized using neural networks and genetic algorithms, respectively. This paper utilizes the design of experiments (DOE) in contact formation to reduce process time and fabrication costs. The experiments were designed by using central composite design which consists of 24 factorial design augmented by 8 axial points with three center points. After contact formation process, the efficiency of the fabricated solar cell is modeled using neural networks. Established efficiency model is then used for the analysis of the process characteristics and process optimization for more efficient solar cell fabrication.

• International Journal of Ocean System Engineering
• /
• v.3 no.3
• /
• pp.116-130
• /
• 2013

The prediction of ocean currents in real time over the warning times of a few hours or days is required in planning many operation-related activities in the ocean. Traditionally this is done through numerical models which are targeted toward producing spatially distributed information. This paper discusses a complementary method to do so when site-specific predictions are desired. It is based on the use of a recurrent type of neural network as well as the statistical tool of model tree. The measurements made at a site in Indian Ocean over a period of 4 years were used. The predictions were made over 72 time steps in advance. The models developed were found to be fairly accurate in terms of the selected error statistics. Among the two modeling techniques the model tree performed better showing the necessity of using distributed models for different sub-domains of data rather than a unique one over the entire input domain. Typically such predictions were associated with average errors of less than 2.0 cm/s. Although the prediction accuracy declined over longer intervals, it was still very satisfactory in terms of theselected error criteria. Similarly prediction of extreme values matched with that of the rest of predictions. Unlike past studies both east-west and north-south current components were predicted fairly well.

• Journal of Biomedical Engineering Research
• /
• v.23 no.3
• /
• pp.197-205
• /
• 2002

The objectives of this Paper is to implement a diagnostic classifier of differential laryngeal diseases from acoustic signals acquired in a noisy room. For this Purpose, the voice signals of the vowel /a/ were collected from Patients in a soundproof chamber and got mixed with noise. Then, the acoustic Parameters were analyzed, and hierarchical neural networks were applied to the data classification. The classifier had a structure of five-step hierarchical neural networks. The first neural network classified the group into normal and benign or malign laryngeal disease cases. The second network classified the group into normal or benign laryngeal disease cases The following network distinguished polyp. nodule. Palsy from the benign laryngeal cases. Glottic cancer cases were discriminated into T1, T2. T3, T4 by the fourth and fifth networks All the neural networks were based on multilayer perceptron model which classified non-linear Patterns effectively and learned by an error back-propagation algorithm. We chose some acoustic Parameters for classification by investigating the distribution of laryngeal diseases and Pilot classification results of those Parameters derived from MDVP. The classifier was tested by using the chosen parameters to find the optimum ones. Then the networks were improved by including such Pre-Processing steps as linear and z-score transformation. Results showed that 90% of T1, 100% of T2-4 were correctly distinguished. On the other hand. 88.23% of vocal Polyps, 100% of normal cases. vocal nodules. and vocal cord Paralysis were classified from the data collected in a noisy room.

• Journal of information and communication convergence engineering
• /
• v.2 no.2
• /
• pp.110-115
• /
• 2004

The prosthesis of current commercialized apparatus has considerable problems, requiring improvement. Especially, LLP(Lower Limb Prosthesis)-related problems have improved, but it cannot provide normal walking because, mainly, the gait control of the LLP does not fit with patient's gait manner. To solve this problem, HCI((Human Computer Interaction) that adapts and controls LLP postures according to patient's gait manner more effectively is studied in this research. The proposed control technique has 2 steps： 1) the multilayer neural network forecasts angles of gait of LLP by using the angle of normal side of lower limbs； and 2) the adaptive neural controller manages the postures of the LLP based on the predicted joint angles. According to the experiment data, the prediction error of hip angles was 0.32［deg.］, and the predicted error of knee angles was 0.12［deg.] for the estimated posture angles for the LLP. The performance data was obtained by applying the reference inputs of the LLP controller while walking. Accordingly, the control performance of the hip prosthesis improved by 80% due to the control postures of the LLP using the reference input when comparing with LQR controller.

• Journal of Korea Multimedia Society
• /
• v.25 no.11
• /
• pp.1653-1671
• /
• 2022

Crop diseases affect crop production, more than 30 billion USD globally. We proposed a classification study of crop species and diseases using deep learning algorithms for corn, cucumber, pepper, and strawberry. Our study has three steps of species classification, disease detection, and disease classification, which is noteworthy for using captured images without additional processes. We designed deep learning approach of deep learning convolutional neural networks based on Mask R-CNN model to classify crop species. Inception and Resnet models were presented for disease detection and classification sequentially. For classification, we trained Mask R-CNN network and achieved loss value of 0.72 for crop species classification and segmentation. For disease detection, InceptionV3 and ResNet101-V2 models were trained for nodes of crop species on 1,500 images of normal and diseased labels, resulting in the accuracies of 0.984, 0.969, 0.956, and 0.962 for corn, cucumber, pepper, and strawberry by InceptionV3 model with higher accuracy and AUC. For disease classification, InceptionV3 and ResNet 101-V2 models were trained for nodes of crop species on 1,500 images of diseased label, resulting in the accuracies of 0.995 and 0.992 for corn and cucumber by ResNet101 with higher accuracy and AUC whereas 0.940 and 0.988 for pepper and strawberry by Inception.