• Journal of Multimedia Information System
• /
• v.9 no.2
• /
• pp.75-86
• /
• 2022

The existing image quality assessment (IQA) datasets have a small number of samples. Some methods based on transfer learning or data augmentation cannot make good use of image quality-related features. A No Reference (NR)-IQA method based on multi-task training and quality awareness is proposed. First, single or multiple distortion types and levels are imposed on the original image, and different strategies are used to augment different types of distortion datasets. With the idea of weak supervision, we use the Full Reference (FR)-IQA methods to obtain the pseudo-score label of the generated image. Then, we combine the classification information of the distortion type, level, and the information of the image quality score. The ResNet50 network is trained in the pre-train stage on the augmented dataset to obtain more quality-aware pre-training weights. Finally, the fine-tuning stage training is performed on the target IQA dataset using the quality-aware weights to predicate the final prediction score. Various experiments designed on the synthetic distortions and authentic distortions datasets (LIVE, CSIQ, TID2013, LIVEC, KonIQ-10K) prove that the proposed method can utilize the image quality-related features better than the method using only single-task training. The extracted quality-aware features improve the accuracy of the model.

• Journal of the Korean Society of Fisheries and Ocean Technology
• /
• v.48 no.1
• /
• pp.72-81
• /
• 2012

Using artificial neural network (ANN) technique, auction prices for common mackerel were forecasted with the daily total sale and auction price data at the Busan Cooperative Fish Market before introducing Total Allowable Catch (TAC) system, when catch data had no limit in Korea. Virtual input data produced from actual data were used to improve the accuracy of prediction and the suitable neural network was induced for the prediction. We tested 35 networks to be retained 10, and found good performance network with regression ratio of 0.904 and determination coefficient of 0.695. There were significant variations between training and verification errors in this network. Ideally, it should require more training cases to avoid over-learning, which leads to improve performance and makes the results more reliable. And the precision of prediction was improved when environmental factors including physical and biological variables were added. This network for prediction of price and catch was considered to be applicable for other fishes.

• Journal of Navigation and Port Research
• /
• v.48 no.2
• /
• pp.109-115
• /
• 2024

To develop a shipping company insolvency prediction model, we sampled shipping companies that closed between 2005 and 2023. In addition, a closed company and a normal company with similar asset size were selected as a paired sample. For this study, data of a total of 82 companies, including 42 closed companies and 42 general companies, were obtained. These data were randomly divided into a training set (2/3 of data) and a testing set (1/3 of data). Training data were used to develop the model while test data were used to measure the accuracy of the model. In this study, a prediction model for Korean shipping insolvency was developed using financial ratio variables frequently used in previous studies. First, using the LASSO technique, main variables out of 24 independent variables were reduced to 9. Next, we set insolvent companies to 1 and normal companies to 0 and fitted logistic regression, LDA and QDA model. As a result, the accuracy of the prediction model was 82.14% for the QDA model, 78.57% for the logistic regression model, and 75.00% for the LDA model. In addition, variables 'Current ratio', 'Interest expenses to sales', 'Total assets turnover', and 'Operating income to sales' were analyzed as major variables affecting corporate insolvency.

• ETRI Journal
• /
• v.45 no.4
• /
• pp.557-569
• /
• 2023

Traditional deterministic channel modeling is accurate in prediction, but due to its complexity, improving computational efficiency remains a challenge. In an alternative approach, we investigated a multilayer artificial neural network (ANN) to predict large-scale and small-scale channel characteristics in metro tunnels. Simulated high-precision training datasets were obtained by combining measurement campaign with a ray tracing (RT) method in a metro tunnel. Performance on the training data was used to determine the number of hidden layers and neurons of the multilayer ANN. The proposed multilayer ANN performed efficiently (10 s for training; 0.19 ms for prediction), and accurately, with better approximation of the RT data than the single-layer ANN. The root mean square errors (RMSE) of path loss (2.82 dB), root mean square delay spread (0.61 ns), azimuth angle spread (3.06°), and elevation angle spread (1.22°) were impressive. These results demonstrate the superior computing efficiency and model complexity of ANNs.

• Journal of Korean Academy of Nursing
• /
• v.51 no.3
• /
• pp.280-293
• /
• 2021

Purpose: This study aimed to identify the risk factors for diabetic foot ulceration (DFU) to develop and evaluate the performance of a DFU prediction model and nomogram among people with diabetes mellitus (DM). Methods: This unmatched case-control study was conducted with 379 adult patients (118 patients with DM and 261 controls) from four general hospitals in South Korea. Data were collected through a structured questionnaire, foot examination, and review of patients' electronic health records. Multiple logistic regression analysis was performed to build the DFU prediction model and nomogram. Further, their performance was analyzed using the Lemeshow-Hosmer test, concordance statistic (C-statistic), and sensitivity/specificity analyses in training and test samples. Results: The prediction model was based on risk factors including previous foot ulcer or amputation, peripheral vascular disease, peripheral neuropathy, current smoking, and chronic kidney disease. The calibration of the DFU nomogram was appropriate (χ² = 5.85, p = .321). The C-statistic of the DFU nomogram was .95 (95% confidence interval .93~.97) for both the training and test samples. For clinical usefulness, the sensitivity and specificity obtained were 88.5% and 85.7%, respectively at 110 points in the training sample. The performance of the nomogram was better in male patients or those having DM for more than 10 years. Conclusion: The nomogram of the DFU prediction model shows good performance, and is thereby recommended for monitoring the risk of DFU and preventing the occurrence of DFU in people with DM.

• Nuclear Engineering and Technology
• /
• v.53 no.6
• /
• pp.1796-1809
• /
• 2021

The subchannel of a research reactor used to generate high power density is designed to be narrow and rectangular and comprises plate-type fuels operating under downward flow conditions. Critical heat flux (CHF) is a crucial parameter for estimating the safety of a nuclear fuel; hence, this parameter should be accurately predicted. Here, machine learning is applied for the prediction of CHF in a narrow rectangular channel. Although machine learning can effectively analyze large amounts of complex data, its application to CHF, particularly for narrow rectangular channels, remains challenging because of the limited flow conditions available in existing experimental databases. To resolve this problem, we used four CHF correlations to generate pseudo-data for training an artificial neural network. We also propose a network architecture that includes pre-training and prediction stages to predict and analyze the CHF. The trained neural network predicted the CHF with an average error of 3.65% and a root-mean-square error of 17.17% for the test pseudo-data; the respective errors of 0.9% and 26.4% for the experimental data were not considered during training. Finally, machine learning was applied to quantitatively investigate the parametric effect on the CHF in narrow rectangular channels under downward flow conditions.

• Journal of the military operations research society of Korea
• /
• v.34 no.1
• /
• pp.1-11
• /
• 2008

A methodology for the improvement of simulation based training system for the artillery is proposed in this paper. The complex nonlinear relationship inherent among parameters in artillery firing is difficult to model and analyze. By introducing neural network based simulation, accurate representation of artillery firing is made possible. The artillery training system can greatly benefit from the improved prediction. Neural networks learning is conducted using the conjugate gradient algorithm. The evaluation of the proposed methodology is performed through simulation. Prediction errors of both regression analysis model and neural networks model are analyzed. Implementation of neural networks to training system enables more realistic training, improved combat power and reduced budget.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.7
• /
• pp.2321-2338
• /
• 2021

Compared with vehicle trajectories, pedestrian trajectories have stronger degrees of freedom and complexity, which poses a higher challenge to trajectory prediction tasks. This paper designs a mode to divide the trajectory of pedestrians at a traffic intersection, which converts the trajectory regression problem into a trajectory classification problem. This paper builds a deep model for pedestrian trajectory prediction at intersections for the task of pedestrian short-term trajectory prediction. The model calculates the spatial correlation and temporal dependence of the trajectory. More importantly, it captures the interactive features among pedestrians through the Attention mechanism. In order to improve the training speed, the model is composed of pure convolutional networks. This design overcomes the single-step calculation mode of the traditional recurrent neural network. The experiment uses Vulnerable Road Users trajectory dataset for related modeling and evaluation work. Compared with the existing models of pedestrian trajectory prediction, the model proposed in this paper has advantages in terms of evaluation indicators, training speed and the number of model parameters.

• Proceedings of the KSRS Conference
• /
• v.2
• /
• pp.812-815
• /
• 2006

Proteins are essential agents for controlling, effecting and modulating cellular functions, and proteins with similar sequences have diverged from a common ancestral gene, and have similar structures and functions. Function prediction of unknown proteins remains one of the most challenging problems in bioinformatics. Recently, various computational approaches have been developed for identification of short sequences that are conserved within a family of closely related protein sequence. Protein function is often correlated with highly conserved motifs. Motif is the smallest unit of protein structure and function, and intends to make core part among protein structural and functional components. Therefore, prediction methods using data mining or machine learning have been developed. In this paper, we describe an approach for protein function prediction of motif-based models using data mining. Our work consists of three phrases. We make training and test data set and construct classifier using a training set. Also, through experiments, we evaluate our classifier with other classifiers in point of the accuracy of resulting classification.

• Transactions of Materials Processing
• /
• v.29 no.4
• /
• pp.218-228
• /
• 2020

The prediction of final mass and optimized process conditions of injection molded products using Artificial Neural Network (ANN) were demonstrated. The ANN was modeled with 10 input parameters and one output parameter (mass). The input parameters, i.e.; melt temperature, mold temperature, injection speed, packing pressure, packing time, cooling time, back pressure, plastification speed, V/P switchover, and suck back were selected. To generate training data for the ANN model, 77 experiments based on the combination of orthogonal sampling and random sampling were performed. The collected training data were normalized to eliminate scale differences between factors to improve the prediction performance of the ANN model. Grid search and random search method were used to find the optimized hyper-parameter of the ANN model. After the training of ANN model, optimized process conditions that satisfied the target mass of 41.14 g were predicted. The predicted process conditions were verified through actual injection molding experiments. Through the verification, it was found that the average deviation in the optimized conditions was 0.15±0.07 g. This value confirms that our proposed procedure can successfully predict the optimized process conditions for the target mass of injection molded products.