• Korean Journal of Cognitive Science
• /
• v.34 no.3
• /
• pp.197-226
• /
• 2023

Generative Text Summarization is one of the Natural Language Processing tasks. It generates a short abbreviated summary while preserving the content of the long text. ROUGE is a widely used lexical-overlap based metric for text summarization models in generative summarization benchmarks. Although it shows very high performance, the studies report that 30% of the generated summary and the text are still inconsistent. This paper proposes a methodology for evaluating the performance of the summary model without using the correct summary. AggreFACT is a human-annotated dataset that classifies the types of errors in neural text summarization models. Among all the test candidates, the two cases, generation summary, and when errors occurred throughout the summary showed the highest correlation results. We observed that the proposed evaluation score showed a high correlation with models finetuned with BART and PEGASUS, which is pretrained with a large-scale Transformer structure.

• Korean Journal of Radiology
• /
• v.22 no.12
• /
• pp.2017-2025
• /
• 2021

Objective: To evaluate the accuracy and clinical efficacy of a hybrid Greulich-Pyle (GP) and modified Tanner-Whitehouse (TW) artificial intelligence (AI) model for bone age assessment. Materials and Methods: A deep learning-based model was trained on an open dataset of multiple ethnicities. A total of 102 hand radiographs (51 male and 51 female; mean age ± standard deviation = 10.95 ± 2.37 years) from a single institution were selected for external validation. Three human experts performed bone age assessments based on the GP atlas to develop a reference standard. Two study radiologists performed bone age assessments with and without AI model assistance in two separate sessions, for which the reading time was recorded. The performance of the AI software was assessed by comparing the mean absolute difference between the AI-calculated bone age and the reference standard. The reading time was compared between reading with and without AI using a paired t test. Furthermore, the reliability between the two study radiologists' bone age assessments was assessed using intraclass correlation coefficients (ICCs), and the results were compared between reading with and without AI. Results: The bone ages assessed by the experts and the AI model were not significantly different (11.39 ± 2.74 years and 11.35 ± 2.76 years, respectively, p = 0.31). The mean absolute difference was 0.39 years (95% confidence interval, 0.33-0.45 years) between the automated AI assessment and the reference standard. The mean reading time of the two study radiologists was reduced from 54.29 to 35.37 seconds with AI model assistance (p < 0.001). The ICC of the two study radiologists slightly increased with AI model assistance (from 0.945 to 0.990). Conclusion: The proposed AI model was accurate for assessing bone age. Furthermore, this model appeared to enhance the clinical efficacy by reducing the reading time and improving the inter-observer reliability.

• Journal of the Korea Convergence Society
• /
• v.13 no.3
• /
• pp.77-89
• /
• 2022

It has been proven through many previous researches that the pretrained language model with a large corpus helps improve performance in various natural language processing tasks. However, there is a limit to building a large-capacity corpus for training in a language environment where resources are scarce. Using the Cross-lingual Post-Training (XPT) method, we analyze the method's efficiency in Korean, which is a low resource language. XPT selectively reuses the English pretrained language model parameters, which is a high resource and uses an adaptation layer to learn the relationship between the two languages. This confirmed that only a small amount of the target language dataset in the relationship extraction shows better performance than the target pretrained language model. In addition, we analyze the characteristics of each model on the Korean language model and the Korean multilingual model disclosed by domestic and foreign researchers and companies.

• Journal of Internet Computing and Services
• /
• v.21 no.3
• /
• pp.103-111
• /
• 2020

Liver cancer is the most fatal cancer that occurs worldwide. In order to diagnose liver cancer, the patient's physical condition was checked by using a CT technique using radiation. Segmentation was needed to diagnose the liver on the patient's abdominal CT scan, which the radiologists had to do manually, which caused tremendous time and human mistakes. In order to automate, researchers attempted segmentation using image segmentation algorithms in computer vision field, but it was still time-consuming because of the interactive based and the setting value. To reduce time and to get more accurate segmentation, researchers have begun to attempt to segment the liver in CT images using CNNs, which show significant performance in various computer vision fields. The pixel value, or numerical value, of the CT image is called the Hounsfield Unit (HU) value, which is a relative representation of the transmittance of radiation, and usually ranges from about -2000 to 2000. In general, deep learning researchers reduce or limit this range and use it for training to remove noise and focus on the target organ. Here, we observed that the range of HU values was limited in many studies but different in various liver segmentation studies, and assumed that performance could vary depending on the HU range. In this paper, we propose the possibility of considering HU value range as a hyper parameter. U-Net and ResUNet were used to compare and experiment with different HU range limit preprocessing of CHAOS dataset under limited conditions. As a result, it was confirmed that the results are different depending on the HU range. This proves that the range limiting the HU value itself can be a hyper parameter, which means that there are HU ranges that can provide optimal performance for various models.

• Journal of Broadcast Engineering
• /
• v.22 no.6
• /
• pp.693-701
• /
• 2017

In this paper, we propose a new method for music genre classification using spikegram and deep neural network. The human auditory system encodes the input sound in the time and frequency domain in order to maximize the amount of sound information delivered to the brain using minimum energy and resource. Spikegram is a method of analyzing waveform based on the encoding function of auditory system. In the proposed method, we analyze the signal using spikegram and extract a feature vector composed of key information for the genre classification, which is to be used as the input to the neural network. We measure the performance of music genre classification using the GTZAN dataset consisting of 10 music genres, and confirm that the proposed method provides good performance using a low-dimensional feature vector, compared to the current state-of-the-art methods.

• Proceedings of the KSME Conference
• /
• 2007.05a
• /
• pp.372-377
• /
• 2007

For more than 2,500 years, surgical teaching has been based on the so called "see one, do one, teach one" paradigm, in which the surgical trainee learns by operating on patients under close supervision of peers and superiors. However, higher demands on the quality of patient care and rising malpractice costs have made it increasingly risky to train on patients. Minimally invasive surgery, in particular, has made it more difficult for an instructor to demonstrate the required manual skills. It has been recognized that, similar to flight simulators for pilots, virtual reality (VR) based surgical simulators promise a safer and more comprehensive way to train manual skills of medical personnel in general and surgeons in particular. One of the major challenges in the development of VR-based surgical trainers is the real-time and realistic simulation of interactions between surgical instruments and biological tissues. It involves multi-disciplinary research areas including soft tissue mechanical behavior, tool-tissue contact mechanics, computer haptics, computer graphics and robotics integrated into VR-based training systems. The research described in this paper addresses the problem of characterizing soft tissue properties for medical virtual environments. A system to measure in vivo mechanical properties of soft tissues was designed, and eleven sets of animal experiments were performed to measure in vivo and in vitro biomechanical properties of porcine intra-abdominal organs. Viscoelastic tissue parameters were then extracted by matching finite element model predictions with the empirical data. Finally, the tissue parameters were combined with geometric organ models segmented from the Visible Human Dataset and integrated into a minimally invasive surgical simulation system consisting of haptic interface devices and a graphic display.

• KIISE Transactions on Computing Practices
• /
• v.24 no.3
• /
• pp.138-144
• /
• 2018

A video is a vivid medium similar to human's visual-linguistic experiences, since it can inculcate a sequence of situations, actions or dialogues that can be told as a story. In this study, we propose story learning/regeneration frameworks from videos with successive event order supervision for contextual coherence. The supervision induces each episode to have a form of trajectory in the latent space, which constructs a composite representation of ordering and semantics. In this study, we incorporated the use of kids videos as a training data. Some of the advantages associated with the kids videos include omnibus style, simple/explicit storyline in short, chronological narrative order, and relatively limited number of characters and spatial environments. We build the encoder-decoder structure with successive event order embedding, and train bi-directional LSTMs as sequence models considering multi-step sequence prediction. Using a series of approximately 200 episodes of kids videos named 'Pororo the Little Penguin', we give empirical results for story regeneration tasks and SEOE. In addition, each episode shows a trajectory-like shape on the latent space of the model, which gives the geometric information for the sequence models.

• International Journal of Human Ecology
• /
• v.12 no.1
• /
• pp.87-99
• /
• 2011

Purpose- The purpose of this study was to compare the fit of two prototype liquid cooled vests using a 3D body scanner and accompanying software. The objectives of this study were to obtain quantitative measurements of ease values, and to use these data to evaluate the fit of two cooling vests in active positions and to develop methodological protocol to resolve alignment issues between the scans using software designed for the alignment of 3D objects. Design/methodology/approach- Garment treatments and body positions were two independent variables with three levels each. Quantitative dataset were dependent variables, and were manipulated in 3x3 factorial designs with repeated measures. Scan images from eight subjects were used and ease values were obtained to compare the fit. Two different types of analyses were conducted in order to compare the fit using t-test; those were radial mean distance value analysis and radial distance distribution rate analysis. Findings- Overall prototype II achieved a closer fit than prototype I with both analyses. These were consistent results with findings from a previous study that used a different approach for evaluation. Research limitations/implications- The main findings can be used as practical feedback for prototype modification/selection in the design process, making use of 3D body scanner as an evaluation tool. Originality/value- Methodological protocols that were devised to eliminate potential sources of errors can contribute to application of data from 3D body scanners.

• Smart Media Journal
• /
• v.5 no.1
• /
• pp.38-43
• /
• 2016

As smart environment is spread out in our living environments, the needs of an approach related to Human Computer Interaction(HCI) is increases. One of them is head pose estimation. it related to gaze direction estimation, since head has a close relationship to eyes by the body structure. It's a key factor in identifying person's intention or the target of interest, hence it is an essential research in HCI. In this paper, we propose an approach for head pose estimation with pre-defined several directions by random forest classifier. We use canny edge detector to extract feature of the different facial image which is obtained between input image and averaged frontal facial image for extraction of rotation information of input image. From that, we obtain the binary edge image, and make two accumulated histograms which are obtained by counting the number of pixel which has non-zero value along each of the axes. This two accumulated histograms are used to feature of the facial image. We use CAS-PEAL-R1 Dataset for training and testing to random forest classifier, and obtained 80.6% accuracy.

• Restorative Dentistry and Endodontics
• /
• v.42 no.1
• /
• pp.27-33
• /
• 2017

Objectives: To determine the optimal timing for post space preparation of root canals sealed with epoxy resin-based AH Plus sealer in terms of its polymerization and influence on apical leakage. Materials and Methods: The epoxy polymerization of AH Plus (Dentsply DeTrey) as a function of time after mixing (8, 24, and 72 hours, and 1 week) was evaluated using Fourier transform infrared (FTIR) spectroscopy and microhardness measurements. The change in the glass transition temperature ($T_g$) of the material with time was also investigated using differential scanning calorimetry (DSC). Fifty extracted human single-rooted premolars were filled with gutta-percha and AH Plus, and randomly separated into five groups (n = 10) based on post space preparation timing (immediately after root canal obturation and 8, 24, and 72 hours, and 1 week after root canal obturation). The extent of apical leakage (mm) of the five groups was compared using a dye leakage test. Each dataset was statistically analyzed by one-way analysis of variance and Tukey's post hoc test (${\alpha}=0.05$). Results: Continuous epoxy polymerization of the material with time was observed. Although the $T_g$ values of the material gradually increased with time, the specimens presented no clear $T_g$ value at 1 week after mixing. When the post space was prepared 1 week after root canal obturation, the leakage was significantly higher than in the other groups (p < 0.05), among which there was no significant difference in leakage. Conclusions: Poor apical seal was detected when post space preparation was delayed until 1 week after root canal obturation.