• Journal of Intelligence and Information Systems
• /
• v.26 no.2
• /
• pp.79-104
• /
• 2020

Recently, as deep learning has attracted attention, the use of deep learning is being considered as a method for solving problems in various fields. In particular, deep learning is known to have excellent performance when applied to applying unstructured data such as text, sound and images, and many studies have proven its effectiveness. Owing to the remarkable development of text and image deep learning technology, interests in image captioning technology and its application is rapidly increasing. Image captioning is a technique that automatically generates relevant captions for a given image by handling both image comprehension and text generation simultaneously. In spite of the high entry barrier of image captioning that analysts should be able to process both image and text data, image captioning has established itself as one of the key fields in the A.I. research owing to its various applicability. In addition, many researches have been conducted to improve the performance of image captioning in various aspects. Recent researches attempt to create advanced captions that can not only describe an image accurately, but also convey the information contained in the image more sophisticatedly. Despite many recent efforts to improve the performance of image captioning, it is difficult to find any researches to interpret images from the perspective of domain experts in each field not from the perspective of the general public. Even for the same image, the part of interests may differ according to the professional field of the person who has encountered the image. Moreover, the way of interpreting and expressing the image also differs according to the level of expertise. The public tends to recognize the image from a holistic and general perspective, that is, from the perspective of identifying the image's constituent objects and their relationships. On the contrary, the domain experts tend to recognize the image by focusing on some specific elements necessary to interpret the given image based on their expertise. It implies that meaningful parts of an image are mutually different depending on viewers' perspective even for the same image. So, image captioning needs to implement this phenomenon. Therefore, in this study, we propose a method to generate captions specialized in each domain for the image by utilizing the expertise of experts in the corresponding domain. Specifically, after performing pre-training on a large amount of general data, the expertise in the field is transplanted through transfer-learning with a small amount of expertise data. However, simple adaption of transfer learning using expertise data may invoke another type of problems. Simultaneous learning with captions of various characteristics may invoke so-called 'inter-observation interference' problem, which make it difficult to perform pure learning of each characteristic point of view. For learning with vast amount of data, most of this interference is self-purified and has little impact on learning results. On the contrary, in the case of fine-tuning where learning is performed on a small amount of data, the impact of such interference on learning can be relatively large. To solve this problem, therefore, we propose a novel 'Character-Independent Transfer-learning' that performs transfer learning independently for each character. In order to confirm the feasibility of the proposed methodology, we performed experiments utilizing the results of pre-training on MSCOCO dataset which is comprised of 120,000 images and about 600,000 general captions. Additionally, according to the advice of an art therapist, about 300 pairs of 'image / expertise captions' were created, and the data was used for the experiments of expertise transplantation. As a result of the experiment, it was confirmed that the caption generated according to the proposed methodology generates captions from the perspective of implanted expertise whereas the caption generated through learning on general data contains a number of contents irrelevant to expertise interpretation. In this paper, we propose a novel approach of specialized image interpretation. To achieve this goal, we present a method to use transfer learning and generate captions specialized in the specific domain. In the future, by applying the proposed methodology to expertise transplant in various fields, we expected that many researches will be actively conducted to solve the problem of lack of expertise data and to improve performance of image captioning.

• Information Systems Review
• /
• v.7 no.2
• /
• pp.101-115
• /
• 2005

IT companies make a lot of effort to share and utilize the experiences of their members and transform them into organizational knowledge as a competitive core. However they face a dilemma in that they have to spend time and financial resources to perform activities around knowledge management for the long-term gains, while carrying at field-work for making short-term profits. As an initial attempt to tackle this managerial problem, this paper tries to investigate the mechanism of knowledge management in a small IT company in Korea with a synthetic view-point using system dynamics simulation model. It depicts the dynamic behaviors of knowledge management and presents some findings of political leverage. Although this model has to be further replenished, the scheme for the dynamism of knowledge management and the findings presented in the paper could be useful for the decision makers, especially of knowledge-intensive organizations.

• Development and Reproduction
• /
• v.9 no.1
• /
• pp.1-5
• /
• 2005

The hypothalamo-pituitary-gonadal hormone axis is centrally controlled by a complex regulatory network of excitatory and inhibitory signals, that is dormant during infantile and juvenile periods and activated at puberty. The kisspeptins are the peptide products of the KiSS-1 gene and the endogenous agonists for the G protein-coupled receptor 54(GPR54). Although KiSS-1 was initially discovered as a metastasis suppressor gene, a recent evidence suggests the KiSS-1/GPR54 system is a key regulator of the reproductive system. Yet the actual role of the KiSS-1/GPR54 system in the neuroendocrine control of gonadotropin secretion remains largely unexplored, the system could be the first missing link in the reproductive hormonal axis. Central or peripheral administration of kisspeptin stimulates the hypothalamic-pituitary-gonadal axis, increasing circulating gonadotropin levels in rodents, sheep, monkey and human models. These effects appear likely to be mediated via the hypothalamic GnRH neuron system, although kisspeptins may have direct effects on the anterior pituitary gland. The loss of function mutations of the GPR54(GPR54-/-) have been associated with lack of puberty onset and idiopathic hypogonadotropic hypogonadism(IHH). So kisspeptin infusion may provide a novel mechanism for HPG axis manipulation in disorders of the reproductive system.

• Clean Technology
• /
• v.13 no.4
• /
• pp.266-273
• /
• 2007

The adsorption properties of the activated carbon-based adsorbents were studied to remove COS emitted from $SO_2$ catalytic reduction process on the integrated gasification combined cycle (IGCC) system in this work. Transition metal supported catalysts and mixed metal oxide catalysts were used for the $SO_2$ catalytic reduction. The mechanism of COS produced from the $SO_2$ reduction and the COS concentration s according to the reaction temperature were investigated. In this study, an activated carbon and a modified activated carbon doped with KOH were used to remove the very low concentration of COS effectively. The adsorption rate and the breakthrough time of COS were measured by a thermo gravity analyzer (TGA, Cahn Balance) and a fixed bed flow reactor equipped with GC-pulsed flammable photometric detector (PFPD), respectively. It was confirmed that the COS breakthrough time of the activated carbon doped with KOH was longer than that of an activated carbon. In conclusion, the modified-activated carbon having a high surface area showed a high adsorption rate of COS produced from the $SO_2$ reduction.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.38 no.10
• /
• pp.712-720
• /
• 2001

Algorithms from the research of the layout migration were proposed in the paper. These are automatic recognition algorithm for the VLSI devices from it, graph based construction algorithm to maintain the constraints, dependencies, and design rule between the devices, and high speed compaction algorithm to reduce size of the VLSI area and reuse the design with compacted size for the new technology. Also, this paper describes that why proposed algorithms are essential for the era of the SoC (System on a Chip), design reuse, and IP DB, which are the big concerns in these days. In addition to introduce our algorithms, the benchmark showed that our performance is superior by 27 times faster than that of the commercial one, and has better efficiency by 3 times in disk usage.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.46 no.5
• /
• pp.419-426
• /
• 2018

For an orbit transfer in a space exploration mission, a solid or liquid rocket booster is included at the last stage of the launch vehicle. During the orbit transfer, thrust misalignment can cause a severe orbit error. Three axis attitude control or spin stabilization can be implemented to minimize the error. Spin stabilization technique has advantages in structural simplicity and lightness. One of ways to apply the spin stabilization to the payload is to include a spin table system in the launch vehicle. In this paper, effect of the spin table system on separation dynamics of the payload is analyzed. Simple model of the spin table to mimic basic functions is designed and simulation environment is established with the model. Effect of the spin table is tested by evaluating separation dynamics of a payload with and without the spin table. Analysis on tolerance effect of separation spring constant on separation dynamics of a payload is conducted.

• The KIPS Transactions:PartD
• /
• v.17D no.5
• /
• pp.337-346
• /
• 2010

Timing diagram is popularly utilized for the reason of its advantages; it is convenient for timing diagram to describe behavior of system and it is simple for described behaviors to recognize it. Various techniques are needed to test systems described in timing diagram. One of them is a technique to derive the system into a certain condition under which a test case is effective. This paper proposes a technique to automatically generate the test input sequence to reach the condition for systems described in timing diagram. It requires a proper input set which satisfy transition condition restricted by input waveform and timing constraints to generate a test input sequence automatically. To solve the problem, this paper chooses an approach utilizing the linear programming, and solving procedure is as follows: 1) Get a Timing diagram model as an input, and transforms the timing diagram model into a linear programming problem. 2) Solve the linear programming problem using a linear programming tool. 3) Generate test input sequences of a timing diagram model from the solution of linear programming problem. This paper addresses the formal method to drive the linear programming model from a given timing diagram, shows the feasibility of our approach by prove it, and demonstrates the usability of our paper by showing that our implemented tool solves an example of a timing diagram model.

• Journal of Life Science
• /
• v.29 no.12
• /
• pp.1401-1407
• /
• 2019

To elucidate whether or not solvent-tolerant bacteria use anti-oxidative defense systems to defend themselves against toxic solvents, oxidative enzyme activity and total anti-oxidative capacity (T-AOC) were investigated in two tolerant strains of Pseudomonas sp. under toluene stress. The superoxide dismutase (SOD) activities of solvent tolerant BCNU 106 exhibited relatively increased levels at a toluene concentration of 100 mg/l, where those of solvent tolerant BCNU 171 increased at 200 mg/l. A greater than three-fold increase in catalase (CAT) levels was observed at concentrations of 200 and 300 mg/l in BCNU 106, and a two-fold increase was monitored at the same concentrations in BCNU 171. High glutathione S-transferase (GST) levels were also observed in the solvent tolerant bacteria. Higher levels of T-AOC was expressed in the solvent tolerant strains than in the ordinary non-tolerant KACC 10266. The highest plateau of SOD in BCNU 171 was observed at 1 hr of toluene exposure. CAT levels plateaued at 1 hr and 14 hr in BCNU 106 and reached the highest plateau at 3 hr in BCNU 171. The highest peak of T-AOC occurred at 9 hr in BCNU 106, and two high peaks occurred in BCNU 171, at 1 hr and at 9 hr of toluene exposure. The solvent-tolerant bacteria showed active antioxidant responses and could survive under harsh environments, including the presence of solvents, through means of antioxidant defense systems.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.11 no.6
• /
• pp.49-59
• /
• 2012

Recently, various alternatives for safety and efficiency of commercial vehicles have been considered, and one of the new alternatives is the application of a digital tachograph. In Korea, the installation of a digital tachograph to commercial vehicles was regulated from 2011 and Korea Transportation Safety Authority developed e-TAS to analyze the monitoring data from digital tachographs installed in the order of 100 commercial vehicles. This study performs the potential impact analysis of the DTG installation, which includes a trend of dangerous driving, a trend of traffic accidents and cost-effective analysis, a trend of fuel consumption and cost-effective analysis, a cost-effective analysis of social benefits using e-TAS data. Depending on the frequency of dangerous driving, the participants are divided into three groups; high-dangerous group, average-dangerous group and low-dangerous group. The high-dangerous driving group shows lower km/liter than the low-dangerous driving group by 15% for buses and taxis and by 30% for trucks. About $CO_2$ emission, the difference becomes bigger; 25%, 25% and 42% for buses, taxis and trucks, respectively. Although this study is a short-term period analysis, the methodology will be applicable for the long-term period analysis with larger data.

• Journal of Broadcast Engineering
• /
• v.27 no.2
• /
• pp.228-239
• /
• 2022

In this paper, we propose an unmanned vehicle scratch detection deep learning model for car sharing services. Conventional scratch detection models consist of two steps: 1) a deep learning module for scratch detection of images before and after rental, 2) a manual matching process for finding newly generated scratches. In order to build a fully automatic scratch detection model, we propose a one-step unmanned scratch detection deep learning model. The proposed model is implemented by applying transfer learning and fine-tuning to the deep learning model that detects changes in satellite images. In the proposed car sharing service, specular reflection greatly affects the scratch detection performance since the brightness of the gloss-treated automobile surface is anisotropic and a non-expert user takes a picture with a general camera. In order to reduce detection errors caused by specular reflected light, we propose a preprocessing process for removing specular reflection components. For data taken by mobile phone cameras, the proposed system can provide high matching performance subjectively and objectively. The scores for change detection metrics such as precision, recall, F1, and kappa are 67.90%, 74.56%, 71.08%, and 70.18%, respectively.