• Journal of radiological science and technology
• /
• v.43 no.6
• /
• pp.503-509
• /
• 2020

This study aims to develop a patient-specific radiation exposure dose prediction model based on anthropometric data that can be easily measurable during CT examination, and to be used as basic data for DRL setting and radiation dose management system in the future. In addition, among the machine learning algorithms, the most suitable model for predicting exposure doses is presented. The data used in this study were chest CT scan data, and a data set was constructed based on the data including the patient's anthropometric data. In the pre-processing and sample selection of the data, out of the total number of samples of 250 samples, only chest CT scans were performed without using a contrast agent, and 110 samples including height and weight variables were extracted. Of the 110 samples extracted, 66% was used as a training set, and the remaining 44% were used as a test set for verification. The exposure dose was predicted through random forest, linear regression analysis, and SVM algorithm using Orange version 3.26.0, an open software as a machine learning algorithm. Results Algorithm model prediction accuracy was R^2 0.840 for random forest, R^2 0.969 for linear regression analysis, and R^2 0.189 for SVM. As a result of verifying the prediction rate of the algorithm model, the random forest is the highest with R^2 0.986 of the random forest, R^2 0.973 of the linear regression analysis, and R^2 of 0.204 of the SVM, indicating that the model has the best predictive power.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.2
• /
• pp.503-523
• /
• 2022

Nowadays, many attendance applications utilise biometric techniques such as the face, fingerprint, and iris recognition. Biometrics has become ubiquitous in many sectors. Due to the advancement of deep learning algorithms, the accuracy rate of biometric techniques has been improved tremendously. This paper proposes a web-based attendance system that adopts facial recognition using open-source deep learning pre-trained models. Face recognition procedural steps using web technology and database were explained. The methodology used the required pre-trained weight files embedded in the procedure of face recognition. The face recognition method includes two important processes: registration of face datasets and face matching. The extracted feature vectors were implemented and stored in an online database to create a more dynamic face recognition process. Finally, user testing was conducted, whereby users were asked to perform a series of biometric verification. The testing consists of facial scans from the front, right (30 - 45 degrees) and left (30 - 45 degrees). Reported face recognition results showed an accuracy of 92% with a precision of 100% and recall of 90%.

• Journal of Internet Computing and Services
• /
• v.23 no.4
• /
• pp.73-85
• /
• 2022

In the era of the 4th industrial revolution, where automation and connectivity are maximized with artificial intelligence, the importance of data collection and utilization for model update is increasing. In order to create a model using artificial intelligence technology, it is usually necessary to gather data in one place so that it can be updated, but this can infringe users' privacy. In this paper, we introduce federated learning, a distributed machine learning method that can update models in cooperation without directly sharing distributed stored data, and introduce a study to optimize distributed consensus among participants without an existing server. In addition, we propose a pattern and group-based distributed consensus optimization algorithm that uses an algorithm for generating patterns and groups based on the Kirkman Triple System, and performs parallel updates and communication. This algorithm guarantees more privacy than the existing distributed consensus optimization algorithm and reduces the communication time until the model converges.

• Journal of the Korea Computer Graphics Society
• /
• v.27 no.3
• /
• pp.1-11
• /
• 2021

Slow sync is a photography technique where a user takes an image with long exposure and a camera flash to enlighten the foreground and background. Unlike short exposure with flash and long exposure without flash, slow sync guarantees the bright foreground and background in the dim environment. However, taking a slow sync image with a smartphone is difficult because the smartphone camera has continuous and weak flash and can not turn on flash if the exposure time is long. This paper proposes a deep learning method that input is a short exposure flash image and output is a slow sync image. We present a deep learning network with a weight map for spatially varying enlightenment. We also propose a dataset that consists of smartphone short exposure flash images and slow sync images for supervised learning. We utilize the linearity of a RAW image to synthesize a slow sync image from short exposure flash and long exposure no-flash images. Experimental results show that our method trained with our dataset synthesizes slow sync images effectively.

• The Journal of Korea Robotics Society
• /
• v.16 no.4
• /
• pp.336-344
• /
• 2021

This paper presents a robust deep learning-based human tracking framework in crowded environments. For practical human tracking applications, a target must be robustly tracked even in undetected or overcrowded situations. The proposed framework consists of two parts: robust deep learning-based human detection and tracking while recognizing the aforementioned situations. In the former part, target candidates are detected using Detectron2, which is one of the powerful deep learning tools, and their weights are computed and assigned. Subsequently, a candidate with the highest weight is extracted and is utilized to track the target human using a Kalman filter. If the bounding boxes of the extracted candidate and another candidate are overlapped, it is regarded as a crowded situation. In this situation, the center information of the extracted candidate is compensated using the state estimated prior to the crowded situation. When candidates are not detected from Detectron2, it means that the target is completely occluded and the next state of the target is estimated using the Kalman prediction step only. In two experiments, people wearing the same color clothes and having a similar height roam around the given place by overlapping one another. The average error of the proposed framework was measured and compared with one of the conventional approaches. In the error result, the proposed framework showed its robustness in the crowded environments.

• Journal of Preventive Medicine and Public Health
• /
• v.56 no.1
• /
• pp.41-49
• /
• 2023

Objectives: Rwanda reported a stunting rate of 33% in 2020, decreasing from 38% in 2015; however, stunting remains an issue. Globally, child deaths from malnutrition stand at 45%. The best options for the early detection and treatment of stunting should be made a community policy priority, and health services remain an issue. Hence, this research aimed to develop a model for predicting stunting in Rwandan children. Methods: The Rwanda Demographic and Health Survey 2019-2020 was used as secondary data. Stratified 10-fold cross-validation was used, and different machine learning classifiers were trained to predict stunting status. The prediction models were compared using different metrics, and the best model was chosen. Results: The best model was developed with the gradient boosting classifier algorithm, with a training accuracy of 80.49% based on the performance indicators of several models. Based on a confusion matrix, the test accuracy, sensitivity, specificity, and F1 were calculated, yielding the model's ability to classify stunting cases correctly at 79.33%, identify stunted children accurately at 72.51%, and categorize non-stunted children correctly at 94.49%, with an area under the curve of 0.89. The model found that the mother's height, television, the child's age, province, mother's education, birth weight, and childbirth size were the most important predictors of stunting status. Conclusions: Therefore, machine-learning techniques may be used in Rwanda to construct an accurate model that can detect the early stages of stunting and offer the best predictive attributes to help prevent and control stunting in under five Rwandan children.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.14 no.9
• /
• pp.1979-1986
• /
• 2010

Biological early warning system detects toxicity by looking at behavior of organisms in water. The system uses classifier for judgement about existence and amount of toxicity in water. Boosting algorithm is one of possible application method for improving performance in a classifier. Boosting repetitively change training example set by focusing on difficult examples in basic classifier. As a result, prediction performance is improved for the events which are difficult to classify, but the information contained in the events which can be easily classified are discarded. In this paper, an incremental learning method to overcome this shortcoming is proposed by using the extended data expression. In this algorithm, decision tree classifier define class distribution information using the weight parameter in the extended data expression by exploiting the necessary information not only from the well classified, but also from the weakly classified events. Experimental results show that the new algorithm outperforms the former Learn++ method without using the weight parameter.

• Physical Therapy Korea
• /
• v.18 no.1
• /
• pp.93-103
• /
• 2011

This study was implemented to verify the feasibility of motor function recovery and the appropriate period for therapy. The research began with spinal laminectomy of 40 white rats of Sprague-Dawley breed and induced them spinal crush injury. Following results were obtained by using the modified Tarlov test (MTT), Basso, Beattle, Bresnahan locomotor rating scale (EBB scale) and modified inclined plate test (MIPT). First, the measurement using the MTT confirm that the most severe aggravation and degeneration of functions are observed two days after induced injury, and no sign of neuromotor function recovery. Second, better scores were achieved by open-ground movement group on BBB locomotor rating scale test, and weight-bearing on inclined plate group show better performance on MIPT. Third, both BBB and MIPT scale manifested the peak of motor function recovery during 16th day after the injury and turn into gradual recovery gradient during 16th to 24th. Fourth, the control group showed functional recovery, however, the level of recovery was less significant when compared with group open-ground movement group and weight-bearing on inclined plate group. Hence, it was clearly manifested that the lumbar region of the spinal cord had shown the best performance when its functions were measured after the execution of specific physical training; therefore it indicated the possibility of learning specific task even in damaged lumbar regions. Thus it is expected to come out with better and more effective functional recovery if concentrated physical therapy was applied starting 4 days after the injury till 16 days, which is the period of the most active recovery.

• Journal of KIISE:Software and Applications
• /
• v.31 no.7
• /
• pp.949-958
• /
• 2004

race detection using a boosting based algorithm requires a very large size of face and nonface data. In addition, the fact that there always occurs a need for adding additional training data for better detection rates demands an efficient incremental teaming algorithm. In the design of incremental teaming based classifiers, the final classifier should represent the characteristics of the entire training dataset. Conventional methods have a critical problem in combining intermediate classifiers that weight updates depend solely on the performance of individual dataset. In this paper, for the purpose of application to face detection, we present a new method to combine an intermediate classifier with previously acquired ones in an optimal manner. Our algorithm creates a validation set by incrementally adding sampled instances from each dataset to represent the entire training data. The weight of each classifier is determined based on its performance on the validation set. This approach guarantees that the resulting final classifier is teamed by the entire training dataset. Experimental results show that the classifier trained by the proposed algorithm performs better than by AdaBoost which operates in batch mode, as well as by ${Learn}^{++}$.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.3
• /
• pp.837-852
• /
• 2021

Recurrent neural network (RNN) architectures have been used for language modeling (LM) tasks that require learning long-range word or character sequences. However, the RNN architecture is still suffered from unstable gradients on long-range sequences. To address the issue of long-range sequences, an attention mechanism has been used, showing state-of-the-art (SOTA) performance in all LM tasks. A differentiable neural computer (DNC) is a deep learning architecture using an attention mechanism. The DNC architecture is a neural network augmented with a content-addressable external memory. However, in the write operation, some information unrelated to the input word remains in memory. Moreover, DNCs have been found to perform poorly with low numbers of weight parameters. Therefore, we propose a robust memory deallocation method using a limited retention vector. The limited retention vector determines whether the network increases or decreases its usage of information in external memory according to a threshold. We experimentally evaluate the robustness of a DNC implementing the proposed approach according to the size of the controller and external memory on the enwik8 LM task. When we decreased the number of weight parameters by 32.47%, the proposed DNC showed a low bits-per-character (BPC) degradation of 4.30%, demonstrating the effectiveness of our approach in language modeling tasks.