• Journal of Korea Multimedia Society
• /
• v.25 no.10
• /
• pp.1375-1385
• /
• 2022

In this paper, we propose a semi-supervised domain adaptation solution to deal with practical face recognition (FR) scenarios where a single face image for each target identity (to be recognized) is only available in the training phase. Main goal of the proposed method is to reduce the discrepancy between the target and the source domain face images, which ultimately improves FR performances. The proposed method is based on the Domain Adatation network (DAN) using an MMD loss function to reduce the discrepancy between domains. In order to train more effectively, we develop a novel loss function learning strategy in which MMD loss and cross-entropy loss functions are adopted by using different weights according to the progress of each epoch during the learning. The proposed weight adoptation focuses on the training of the source domain in the initial learning phase to learn facial feature information such as eyes, nose, and mouth. After the initial learning is completed, the resulting feature information is used to training a deep network using the target domain images. To evaluate the effectiveness of the proposed method, FR performances were evaluated with pretrained model trained only with CASIA-webface (source images) and fine-tuned model trained only with FERET's gallery (target images) under the same FR scenarios. The experimental results showed that the proposed semi-supervised domain adaptation can be improved by 24.78% compared to the pre-trained model and 28.42% compared to the fine-tuned model. In addition, the proposed method outperformed other state-of-the-arts domain adaptation approaches by 9.41%.

• Water for future
• /
• v.25 no.3
• /
• pp.65-77
• /
• 1992

The purpose of this study is to analyse existing hysteresis models and to propose a new type of model. The existing hysteresis models are classified by three types: interpolation model, scaling model and domain model, of which the domain model is based on the theoretical approach. Models which need one branch of hysteresis loop for calibration are developed based on the independent domain concept, however, they are not successful to accurately simulate the real data and Rubicon Sandy Loam and Dune Sand. There is a possibility that a new model is based on the dependent domain model considering the pore blockage effect against air entry for homogeneous porous media(modelIII-1, Mualem, 1984). Concludingly, a new type of hysteresis model is proposed by simplifying ModelIII-1 using a proper assumption.

• Korean Journal of Cognitive Science
• /
• v.18 no.2
• /
• pp.179-191
• /
• 2007

In a goal-oriented dialogue, spoken' intentions can be represented by domain actions that consist of pairs of a speech art and a concept sequence. The domain action prediction of user's utterance is useful to correct some errors that occur in a speech recognition process, and the domain action prediction of system's utterance is useful to generate flexible responses. In this paper, we propose a model to predict a domain action of the next utterance using a neural network. The proposed model predicts the next domain action by using a dialogue history vector and a current domain action as inputs of the neural network. In the experiment, the proposed model showed the precision of 80.02% in speech act prediction and the precision of 82.09% in concept sequence prediction.

• Proceedings of the Korea Inteligent Information System Society Conference
• /
• 2001.01a
• /
• pp.410-420
• /
• 2001

A primitive conceptualization is defined as the set of all intended situations. A non-primitive conceptualization is defined as the set of all the pairs every of which consists of an intended knowledge system and the set of all the situations admitted by the knowledge system. The reality of a domain is considered as the set of all the situation which have ever taken place in the past, are taking place now and will take place in the future. A conceptualization is defined as precise if the set of intended situations is equal to the domain reality. The representation of various elements of a domain ontology in a model of the ontology is considered. These elements are terms for situation description and situations themselves, terms for knowledge description and knowledge systems themselves, mathematical terms and constructions, auxiliary terms and ontological agreements. It has been shown that any ontology representing a conceptualization has to be non-primitive if either (1) a conceptualization contains intended situations of different structures, or (2) a conceptualization contains concepts designated by terms for knowledge description, or (3) a conceptualization contains concept classes and determines properties of the concepts belonging to these classes, but the concepts themselves are introduced by domain knowledge, or (4) some restrictions on meanings of terms for situation description in a conceptualization depend on the meaning of terms for knowledge description.

• Journal of Agricultural Extension & Community Development
• /
• v.21 no.4
• /
• pp.1029-1060
• /
• 2014

The purpose of this study are to figure out whether there are the meaningful differences between learner's learning achievements and contents preference in accordance with the delivery strategies (instructor-focused model, learner-focused model) of learning materials suggested by Bloom in web-based instruction, and to suggest the various options on the contents delivery strategies to improve the learner's learning achievements of each learning domains. Learning domains were divided by the cognitive domain, the affective domain, and the psychomotor domain. The result of research with 182 learners showed that learner-focused model in the cognitive domain caused higher learning achievements and preference than instructor-focused model. And instructor-focused model in the psychomotor domain compared with learner-focused model caused higher learning achievements and preference. However, there were less meaningful differences in the affective domain. In other words, learner-focused model is appropriate to the feature of the cognitive domain while instructor-focused model is appropriate to the feature of the psychomotor domain. The results suggest that delivery strategies should be chosen by domains of learning contents in order to improve learner's learning achievements in web-based instruction. Learner-focused delivery strategies in the cognitive domain and instructor-focused delivery strategies in the psychomotor domain need to be considered positively. Delivery strategies should be studied and developed in order to lead higher learning achievements and preference.

• Korean Journal of Hydrosciences
• /
• v.4
• /
• pp.33-49
• /
• 1993

The Purpose of this study is to classify existing hysteresis models and to discuss a possibility of a new type of the hysteresis model. The existing hysteresis models are classified into three types: the interpolation model, the scaling model and the domain model, of which only domain model is to simulate hysteresis curves based on the theoretical approach, It is useful to develop a hysteresis model that requires only one branch of hysteresis curves for the model calibration because obtaining hysteresis curves by experiments is expensive and time-concept by many investigators, however their models are not successful to accurately simulate real data of Rubicon Sandy Loam and Dune Sand. There is a possibility that a new model is based on the dependent domain concept considering the weighting factor, $P_a$($\theta$), which accounts for the pore blockage effect against air entry. Conclusively, a new model where the weighting factor $P_a$($\theta$) in Model III-1 (Mualem, 1984) reduces to a known variable through an appropriate method is an alternative model which required only one branch of main curves for the model calibration.

• Journal of Information Processing Systems
• /
• v.13 no.1
• /
• pp.57-67
• /
• 2017

In this paper, Atanassov's intuitionistic fuzzy set theory is used to handle the uncertainty of students' knowledgeon domain concepts in an E-learning system. Their knowledge on these domain concepts has been collected from tests that were conducted during their learning phase. Atanassov's intuitionistic fuzzy user model is proposed to deal with vagueness in the user's knowledge description in domain concepts. The user model uses Atanassov's intuitionistic fuzzy sets for knowledge representation and linguistic rules for updating the user model. The scores obtained by each student were collected in this model and the decision about the students' knowledge acquisition for each concept whether completely learned, completely known, partially known or completely unknown were placed into the information table. Finally, it has been found that the proposed scheme is more appropriate than the fuzzy scheme.

• Atmosphere
• /
• v.31 no.1
• /
• pp.17-28
• /
• 2021

We investigated the impact of domain size on the simulated summer precipitation over the Korean Peninsula using the Weather Research and Forecasting (WRF) model. Two different domains are integrated up to 72-hours from 29 June 2017 to 28 July 2017 when the Changma front is active. The domain sizes are adopted from previous RDAPS (Regional Data Assimilation and Prediction System) and current LDAPS (Local Data Assimilation and Prediction System) operated by the Korea Meteorological Administration, while other model configurations are fixed identically. We found that the larger domain size showed better prediction skills, especially in precipitation forecast performance. This performance improvement is particularly noticeable over the central region of the Korean Peninsula. Comparisons of physical aspects of each variable revealed that the inflow of moisture flux from the East China Sea was well reproduced in the experiment with a large model domain due to a more realistic North Pacific high compared to the small domain experiment. These results suggest that the North Pacific anticyclone could be an important factor for the precipitation forecast during the summer-time over the Korean Peninsula.

• Geophysics and Geophysical Exploration
• /
• v.24 no.4
• /
• pp.158-163
• /
• 2021

The frequency-domain induced polarization (IP) response based on Cole-Cole model is expressed as a simple equation in close form. However, it is difficult to compute the time-domain IP response based on Cole-Cole model or any other relaxation model because it cannot be written in closed form. In this study, using numerical experiments, we compared three numerical methods for calculating the time-domain IP response of the Cole-Cole model asymptotically: series expansion, digital linear filtering and Fourier transform. The series expansion method is inadequately accurate for certain time values and converges very slowly. A digital linear filter specially designed to calculate the time-domain IP response does not present the desired accuracy, especially at later times. The Fourier transform method can overcome the abovementioned problems and present the time-domain IP response with adequate accuracy for all time values, even though more computing time is required.

• 한국지구물리탐사학회:학술대회논문집
• /
• 2003.11a
• /
• pp.464-469
• /
• 2003

The traditional and still most widely used, test methods for concrete structures are destructive method, such as coring, drilling or otherwise removing part of the structure to permit visual inspection of the interior. While these methods are highly reliable, they are also time consuming and expensive, and the defects they leave behind often become focal point for deterioration. In this study, tomography by theoretical inversion method in case of elastic wave using impact-echo method among concrete non-destruction test method was made. Taken model experiments are theoretical inversion method and time domain and frequency domain test on pier test model at laboratory level. Also experiment concerning frequency domain on 3 kinds of tunnel model with I-dimension form was carried out.