• Journal of Broadcast Engineering
• /
• v.27 no.7
• /
• pp.1011-1020
• /
• 2022

Recent high-performance panoptic segmentation models are based on transformer architectures. However, transformer-based panoptic segmentation methods are basically slower than convolution-based methods, since the attention mechanism in the transformer requires quadratic complexity w.r.t. image resolution. Also, sine and cosine computation for positional embedding in the transformer also yields a bottleneck for computation time. To address these problems, we adopt three modules to speed up the inference runtime of the transformer-based panoptic segmentation. First, we perform channel-level reduction using depth-wise separable convolution for inputs of the transformer decoder. Second, we replace sine and cosine-based positional encoding with convolution operations, called conv-embedding. We also apply a separable self-attention to the transformer encoder to lower quadratic complexity to linear one for numbers of image pixels. As result, the proposed model achieves 44% faster frame per second than baseline on ADE20K panoptic validation dataset, when we use all three modules.

• ETRI Journal
• /
• v.45 no.1
• /
• pp.93-104
• /
• 2023

Coreference resolution is a task in discourse analysis that links several headwords used in any document object. We suggest pointer networks-based coreference resolution for Korean using multi-task learning (MTL) with an attention mechanism for a hierarchical structure. As Korean is a head-final language, the head can easily be found. Our model learns the distribution by referring to the same entity position and utilizes a pointer network to conduct coreference resolution depending on the input headword. As the input is a document, the input sequence is very long. Thus, the core idea is to learn the word- and sentence-level distributions in parallel with MTL, while using a shared representation to address the long sequence problem. The suggested technique is used to generate word representations for Korean based on contextual information using pre-trained language models for Korean. In the same experimental conditions, our model performed roughly 1.8% better on CoNLL F1 than previous research without hierarchical structure.

• Korean Journal of Health Psychology
• /
• v.16 no.3
• /
• pp.557-575
• /
• 2011

This study tried to examine the characteristics of attention deficits in patients with Obstructive Sleep Apenea(OSA) with different age levels, and to examine which indices of polysomnograms might be related to the indices of attention deficits in OSAs. Two age-level groups and a normal control group were subjected to two computerized attention tests, including a continuous performance test(CPT) and a change blindness task(CBT). In addition, the three groups were subjected to a Polysomnography to extract several sub-indicators of polysomnogram, and an Epworth Sleepiness Scale which measures subjective sleepiness. As results, the OSAs showed significantly more omission and commission errors in CPT, and they showed lower accuracy in CBT compared to the normal group. The results of a correlational analysis showed that attention deficits in OSA are significantly correlated with arterial oxygen saturation among sub-indicators of polysomnograms. In conclusion, OSAs seems to be less attentive, having difficulties in response inhibition, and having deficiencies in noticing important environmental changes. Age seems to make these deficiencies even worse. Especially, the relationship between attention deficiency and hypoxia which could cause irreversible cerebrum damage has an implication in cognitive impairment prevention through early treatment.

• Journal of Korea Multimedia Society
• /
• v.14 no.3
• /
• pp.378-391
• /
• 2011

In this paper, we suggest a novel method of selecting visual information based on bottom-up visual attention of human. We propose a new model that improve accuracy of detecting attention region by using depth information in addition to low-level spatial features such as color, lightness, orientation, form and temporal feature such as motion. Motion is important cue when we derive temporal saliency. But noise obtained during the input and computation process deteriorates accuracy of temporal saliency Our system exploited the result of psychological studies in order to remove the noise from motion information. Although typical systems get problems in determining the saliency if several salient regions are partially occluded and/or have almost equal saliency, our system is able to separate the regions with high accuracy. Spatiotemporally separated prominent regions in the first stage are prioritized using depth value one by one in the second stage. Experiment result shows that our system can describe the salient regions with higher accuracy than the previous approaches do.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.8
• /
• pp.3168-3186
• /
• 2020

Automatic topic essay generation (TEG) is a controllable text generation task that aims to generate informative, diverse, and topic-consistent essays based on multiple topics. To make the generated essays of high quality, a reasonable method should consider both diversity and topic-consistency. Another essential issue is the intrinsic link of the topics, which contributes to making the essays closely surround the semantics of provided topics. However, it remains challenging for TEG to fill the semantic gap between source topic words and target output, and a more powerful model is needed to capture the semantics of given topics. To this end, we propose a pretraining-based contextual self-attention (PC-SAN) model that is built upon the seq2seq framework. For the encoder of our model, we employ a dynamic weight sum of layers from BERT to fully utilize the semantics of topics, which is of great help to fill the gap and improve the quality of the generated essays. In the decoding phase, we also transform the target-side contextual history information into the query layers to alleviate the lack of context in typical self-attention networks (SANs). Experimental results on large-scale paragraph-level Chinese corpora verify that our model is capable of generating diverse, topic-consistent text and essentially makes improvements as compare to strong baselines. Furthermore, extensive analysis validates the effectiveness of contextual embeddings from BERT and contextual history information in SANs.

• Physical Therapy Rehabilitation Science
• /
• v.8 no.3
• /
• pp.158-161
• /
• 2019

Objective: Attentional focus is one of the critical factors that has consistently been demonstrated to enhance motor performance and motor skill. Focusing attention on the inside of the body while engaging in a particular exercise is called internal focus (IF) and focus on the external environment is called external focus (EF). The purpose of this study was to identify effects of IF and EF of attention on postural balance in healthy school-aged children. Design: Cross-sectional study. Methods: Twenty-four healthy school-aged children participated in this study. School-aged children was defined as children ages 8-12 years old. They performed the one-legged standing with EF (focusing on the marker at the level of participants' chest and 150 cm away), IF (focusing the supporting feet), and control (no instruction) respectively. The order of the focus condition was randomly selected. The center of pressure (COP) range, distance, and velocity was measured to compare the effects of applying different attentional focuses in the three conditions. Results: The results of our study show that differences in COP range, distance, and velocity among groups were not significant between the different attentional focuses, although all variables of EF were smaller than IF. It is postulated that the reason for this may be that school school-aged children between 8-12 years old go through a transitional phase from IF to EF in effective motor learning. Conclusions: These findings reveal that the type of attentional focus did not have any effect on postural balance in healthy school-aged children.

• Journal of the Korean Academy of Child and Adolescent Psychiatry
• /
• v.20 no.1
• /
• pp.17-22
• /
• 2009

Objectives : A higher level of aggression has been reported for children with attention-deficit/hyperactivity disorder (ADHD) than for non-ADHD children. Aggression was shown to have a negative effect on the social functioning of children with ADHD. The ability to recognize facial emotion expression has also been related to aggression. In this study, we examined whether impulsivity and dysfunctional recognition of facial emotion expression could explain the aggressiveness of children with ADHD. Methods : 67 children with ADHD participated in this study. We measured the ability to recognize facial emotion expression by using the Emotion Recognition Test (ERT) and we measured aggression by the T score of the aggression subscale of the Child Behavior Checklist (CBCL). Impulsivity was measured by the ADHD diagnostic system (ADS). Results : The teacher rated level of aggression was related to the score of recognizing negative affect. After controlling for the effect of impulsivity, this relationship is not significant. Only the score of the visual commission errors ex plained the level of aggression of children with ADHD. Conclusion : Impulsivity seems to have a major role in explaining the aggression of children with ADHD. The clinical implication of this study is that effective intervention for controlling impulsivity may be expected to reduce the aggression of children with ADHD.

• Knowledge Management Research
• /
• v.14 no.1
• /
• pp.39-55
• /
• 2013

As the importance of intellectual property rights for the 21st century challenge became prominent, many companies have been trying to secure many rights competitively. In particular, application numbers of patent that represents technology has been increased continuously. Korean companies were not exception; mainly in large companies, there have been continued the efforts to grow the number of patent applications in quantitative volume. But the issues that how viable patents the companies have, how effectively the companies manage, and how economically usable the patents are, are totally different from quantitative management level. As such, the issue is connected to how to assess the patent management level of companies. On the other side of quantitative growth of patent that companies hold, there are some problems such as the difficulties to determine if the patents hold substantial values, and the difficulties to determine whether the patent are managed effectively. In addition, as the numbers of patent application and registration of companies are increased, the cost for patent holding and managing increase. It is required to pay continuous attention to the cost of patent management because patent registration fee has a property that increases rapidly with time and burden for patentee become heavier. As a result of analysis and interpretation, we confirmed that quantitative management, particularly the number of patent applicant does not make positive impact on how to use patent after application. Rather, it is observed that the economical usage is influenced positively by the efforts of patent applicants such as considering for the time of patent examination, paying attention to receive patent registration decision. Therefore, this study shows that efforts patent applicants provide in management level after application time are important to maintain the value of patents.

• Journal of the Korea Safety Management & Science
• /
• v.11 no.3
• /
• pp.19-26
• /
• 2009

It is becoming more and more important to develop safety-critical systems with special attention. Examples of the safety-critical systems include the mass transportation systems such as high speed trains, airplanes, ships and so forth. Safety critical issues can also exist in the development of atomic power plants that are attracting a great deal of attention recently as oil prices are sky-rocketing. Note that the safety-critical systems are in general large-scale and very complex for which case the effects of adopting the systems engineering (SE) approach has been quite phenomenal. Furthermore, safety-critical requirements should necessarily be realized in the design phase and be effectively maintained thereafter. In light of these comments, we have considered our approach to developing safety-critical systems to be based on the method combining the systems engineering and safety management processes. To do so, we have developed a design environment by constructing a whole life cycle model in two steps. In the first step, the integrated process model was developed by integrating the SE (ISO/IEC 15283) and systems safety (e.g., hazard analysis) activities and implemented in a computer-aided SE tool environment. The model was represented by three hierarchical levels: the life-cycle level, the process level, and the activity level. As a result, one can see from the model when and how the required SE and safety processes have to be carried out concurrently and iterately. Finally, the design environment was verified by the computer simulation.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.11
• /
• pp.4065-4083
• /
• 2021

Super-resolution can improve the clarity of low-resolution (LR) images, which can increase the accuracy of high-level compute vision tasks. Portable devices have low computing power and storage performance. Large-scale neural network super-resolution methods are not suitable for portable devices. In order to save the computational cost and the number of parameters, Lightweight image processing method can improve the processing speed of portable devices. Therefore, we propose the Enhanced Information Multiple Distillation Network (EIMDN) to adapt lower delay and cost. The EIMDN takes feedback mechanism as the framework and obtains low level features through high level features. Further, we replace the feature extraction convolution operation in Information Multiple Distillation Block (IMDB), with Ghost module, and propose the Enhanced Information Multiple Distillation Block (EIMDB) to reduce the amount of calculation and the number of parameters. Finally, coordinate attention (CA) is used at the end of IMDB and EIMDB to enhance the important information extraction from Spaces and channels. Experimental results show that our proposed can achieve convergence faster with fewer parameters and computation, compared with other lightweight super-resolution methods. Under the condition of higher peak signal-to-noise ratio (PSNR) and higher structural similarity (SSIM), the performance of network reconstruction image texture and target contour is significantly improved.