• The Journal of the Korea institute of electronic communication sciences
• /
• v.16 no.3
• /
• pp.401-406
• /
• 2021

Recently, mean-teacher models based on convolutional recurrent neural networks are popularly used in audio event detection. The mean-teacher model is an architecture that consists of two parallel CRNNs and it is possible to train them effectively on the weakly-labelled and unlabeled audio data by using the consistency learning metric at the output of the two neural networks. In this study, we tried to improve the performance of the mean-teacher model by using additional derivative features of the log-mel spectrum. In the audio event detection experiments using the training and test data from the Task 4 of the DCASE 2018/2019 Challenges, we could obtain maximally a 8.1% relative decrease in the ER(Error Rate) in the mean-teacher model using proposed derivative features.

• Journal of Information Technology Services
• /
• v.18 no.2
• /
• pp.143-159
• /
• 2019

Machine learning (ML) is a method of fitting given data to a mathematical model to derive insights or to predict. In the age of big data, where the amount of available data increases exponentially due to the development of information technology and smart devices, ML shows high prediction performance due to pattern detection without bias. The feature engineering that generates the features that can explain the problem to be solved in the ML process has a great influence on the performance and its importance is continuously emphasized. Despite this importance, however, it is still considered a difficult task as it requires a thorough understanding of the domain characteristics as well as an understanding of source data and the iterative procedure. Therefore, we propose methods to apply deep learning for solving the complexity and difficulty of feature extraction and improving the performance of ML model. Unlike other techniques, the most common reason for the superior performance of deep learning techniques in complex unstructured data processing is that it is possible to extract features from the source data itself. In order to apply these advantages to the business problems, we propose deep learning based methods that can automatically extract features from transaction data or directly predict and classify target variables. In particular, we applied techniques that show high performance in existing text processing based on the structural similarity between transaction data and text data. And we also verified the suitability of each method according to the characteristics of transaction data. Through our study, it is possible not only to search for the possibility of automated feature extraction but also to obtain a benchmark model that shows a certain level of performance before performing the feature extraction task by a human. In addition, it is expected that it will be able to provide guidelines for choosing a suitable deep learning model based on the business problem and the data characteristics.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.1
• /
• pp.24-29
• /
• 2022

Semantic role labeling task used in various natural language processing fields, such as information extraction and question answering systems, is the task of identifying the arugments for a given sentence and predicate. Predicate used as semantic role labeling input are extracted using lexical analysis results such as POS-tagging, but the problem is that predicate can't extract all linguistic patterns because predicate in korean language has various patterns, depending on the meaning of sentence. In this paper, we propose a korean predicate recognition method using neural network model with pre-trained embedding models and lexical features. The experiments compare the performance on the hyper parameters of models and with or without the use of embedding models and lexical features. As a result, we confirm that the performance of the proposed neural network model was 92.63%.

• Tunnel and Underground Space
• /
• v.28 no.6
• /
• pp.670-691
• /
• 2018

This study presents the research results of the BMT(Benchmark Model Test) simulations of the DECOVALEX-2019 project Task B. Task B named 'Fault slip modelling' is aiming at developing a numerical method to predict fault reactivation and the coupled hydro-mechanical behavior of fault. BMT scenario simulations of Task B were conducted to improve each numerical model of participating group by demonstrating the feasibility of reproducing the fault behavior induced by water injection. The BMT simulations consist of seven different conditions depending on injection pressure, fault properties and the hydro-mechanical coupling relations. TOUGH-FLAC simulator was used to reproduce the coupled hydro-mechanical process of fault slip. A coupling module to update the changes in hydrological properties and geometric features of the numerical mesh in the present study. We made modifications to the numerical model developed in Task B Step 1 to consider the changes in compressibility, Permeability and geometric features with hydraulic aperture of fault due to mechanical deformation. The effects of the storativity and transmissivity of the fault on the hydro-mechanical behavior such as the pressure distribution, injection rate, displacement and stress of the fault were examined, and the results of the previous step 1 simulation were updated using the modified numerical model. The simulation results indicate that the developed model can provide a reasonable prediction of the hydro-mechanical behavior related to fault reactivation. The numerical model will be enhanced by continuing interaction and collaboration with other research teams of DECOVALEX-2019 Task B and validated using the field experiment data in a further study.

• The Korean Journal of Archival Studies
• /
• no.32
• /
• pp.127-176
• /
• 2012

Metadata is well recognized as one of the foundational factors in archiving and long-term preservation of digital resources. There are several metadata standards for records management, archives and preservation, e.g. ISAD(G), EAD, AGRkMs, PREMIS, and OAIS. Consideration is important in selecting appropriate metadata standards in order to design metadata schema that meet the requirements of a particular archival system. Interoperability of metadata with other systems should be considered in schema design. In our previous research, we have presented a feature analysis of metadata standards by identifying the primary resource lifecycle stages where each standard is applied. We have clarified that any single metadata standard cannot cover the whole records lifecycle for archiving and preservation. Through this feature analysis, we analyzed the features of metadata in the whole records lifecycle, and we clarified the relationships between the metadata standards and the stages of the lifecycle. In the previous study, more detailed analysis was left for future study. This paper proposes to analyze the metadata schemas from the viewpoint of tasks performed in the lifecycle. Metadata schemas are primarily defined to describe properties of a resource in accordance with the purposes of description, e.g. finding aids, records management, preservation and so forth. In other words, the metadata standards are resource- and purpose-centric, and the resource lifecycle is not explicitly reflected in the standards. There are no systematic methods for mapping between different metadata standards in accordance with the lifecycle. This paper proposes a method for mapping between metadata standards based on the tasks contained in the resource lifecycle. We first propose a Task Model to clarify tasks applied to resources in each stage of the lifecycle. This model is created as a task-centric model to identify features of metadata standards and to create mappings among elements of those standards. It is important to categorize the elements in order to limit the semantic scope of mapping among elements and decrease the number of combinations of elements for mapping. This paper proposes to use 5W1H (Who, What, Why, When, Where, How) model to categorize the elements. 5W1H categories are generally used for describing events, e.g. news articles. As performing a task on a resource causes an event and metadata elements are used in the event, we consider that the 5W1H categories are adequate to categorize the elements. By using these categories, we determine the features of every element of metadata standards which are AGLS, AGRkMS, PREMIS, EAD, OAIS and an attribute set extracted from DPC decision flow. Then, we perform the element mapping between the standards, and find the relationships between the standards. In this study, we defined a set of terms for each of 5W1H categories, which typically appear in the definition of an element, and used those terms to categorize the elements. For example, if the definition of an element includes the terms such as person and organization that mean a subject which contribute to create, modify a resource the element is categorized into the Who category. A single element can be categorized into one or more 5W1H categories. Thus, we categorized every element of the metadata standards using the 5W1H model, and then, we carried out mapping among the elements in each category. We conclude that the Task Model provides a new viewpoint for metadata schemas and is useful to help us understand the features of metadata standards for records management and archives. The 5W1H model, which is defined based on the Task Model, provides us a core set of categories to semantically classify metadata elements from the viewpoint of an event caused by a task.

• Proceedings of the Korea Society of Information Technology Applications Conference
• /
• 2005.11a
• /
• pp.229-232
• /
• 2005

In this paper, an image-based "approach-align -grasp" visual servo control design is proposed for the problem of object grasping, which is based on the binocular stand-alone system. The basic idea consists of considering a vision system as a specific sensor dedicated a task and included in a control servo loop, and we perform automatic grasping follows the classical approach of splitting the task into preparation and execution stages. During the execution stage, once the image-based control modeling is established, the control task can be performed automatically. The proposed visual servoing control scheme ensures the convergence of the image-features to desired trajectories by using the Jacobian matrix, which is proved by the Lyapunov stability theory. And we also stress the importance of projective invariant object/gripper alignment. The alignment between two solids in 3-D projective space can be represented with view-invariant, more precisely; it can be easily mapped into an image set-point without any knowledge about the camera parameters. The main feature of this method is that the accuracy associated with the task to be performed is not affected by discrepancies between the Euclidean setups at preparation and at task execution stages. Then according to the projective alignment, the set point can be computed. The robot gripper will move to the desired position with the image-based control law. In this paper we adopt a constant Jacobian online. Such method describe herein integrate vision system, robotics and automatic control to achieve its goal, it overcomes disadvantages of discrepancies between the different Euclidean setups and proposes control law in binocular-stand vision case. The experimental simulation shows that such image-based approach is effective in performing the precise alignment between the robot end-effector and the object.

• Journal of the Korean earth science society
• /
• v.41 no.1
• /
• pp.61-74
• /
• 2020

The purpose of this study was to compare rock descriptors used by students in two different task contexts and to suggest the characteristics of a task suitable for learning of rocks. Twenty-four 3^rd grade students were given descriptive and inferential tasks about three types of sedimentary rocks, and the rock descriptors used by the students were analyzed from a resources-based view (RBV) about students' conceptions. The result showed that the number of students using everyday descriptors to describe properties of the rocks and the frequency of using the everyday descriptors decreased in the inferential task. It was also revealed that the students using disciplinarily more appropriate descriptors were more likely to infer the process of rock formation in scientifically valid ways. By contrast, student inferences lacking scientific validity were mostly those that used everyday descriptors to express properties of the rocks. Based on these findings, it was concluded that inferential tasks would be suitable for student learning of rocks which is to be authentic to the essential features of earth science practices.

• Journal of the Korea Convergence Society
• /
• v.12 no.7
• /
• pp.37-44
• /
• 2021

Quality estimation (QE) can evaluate the quality of machine translation output even for those who do not know the target language, and its high utilization highlights the need for QE. QE shared task is held every year at Conference on Machine Translation (WMT), and recently, researches applying Pretrained Language Model (PLM) are mainly being conducted. In this paper, we conduct a survey on the QE task and research trends, and we summarize the features of PLM. In addition, we used a multilingual BART model that has not yet been utilized and performed comparative analysis with the existing studies such as XLM, multilingual BERT, and XLM-RoBERTa. As a result of the experiment, we confirmed which PLM was most effective when applied to QE, and saw the possibility of applying the multilingual BART model to the QE task.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.6
• /
• pp.2806-2825
• /
• 2018

Unsupervised learning has shown good performance on image, video and audio classification tasks, and much progress has been made so far. It studies how systems can learn to represent particular input patterns in a way that reflects the statistical structure of the overall collection of input patterns. Many promising deep learning systems are commonly trained by the greedy layerwise unsupervised learning manner. The performance of these deep learning architectures benefits from the unsupervised learning ability to disentangling the abstractions and picking out the useful features. However, the existing unsupervised learning algorithms are often difficult to train partly because of the requirement of extensive hyperparameters. The tuning of these hyperparameters is a laborious task that requires expert knowledge, rules of thumb or extensive search. In this paper, we propose a simple and effective unsupervised feature learning algorithm for image classification, which exploits an explicit optimizing way for population and lifetime sparsity. Firstly, a sparse target matrix is built by the competitive rules. Then, the sparse features are optimized by means of minimizing the Euclidean norm ($L_2$) error between the sparse target and the competitive layer outputs. Finally, a classifier is trained using the obtained sparse features. Experimental results show that the proposed method achieves good performance for image classification, and provides discriminative features that generalize well.

• Proceedings of the Korean Society for Agricultural Machinery Conference
• /
• 1993.10a
• /
• pp.1230-1242
• /
• 1993

In a case of mushroom (Lentinus Edodes L.) , visual features are crucial for grading and the quantitative evaluation of the growth state. The extracted quantitative visual features can be used as a performance index for the drying process control or used for the automatic sorting and grading task. First, primary external features of the front and back sides of mushroom were analyzed. And computer vision based algorithm were developed for the extraction and measurement of those features. An automatic thresholding algorithm , which is the combined type of the window extension and maximum depth finding was developed. Freeman's chain coding was modified by gradually expanding the mask size from 3X3 to 9X9 to preserve the boundary connectivity. According to the side of mushroom determined from the automatic recognition algorithm size thickness, overall shape, and skin texture such as pattern, color (lightness) ,membrane state, and crack were quantified and measured. A portion of t e stalk was also identified and automatically removed , while reconstructing a new boundary using the Overhauser curve formulation . Algorithms applied and developed were coded using MS_C language Ver, 6.0, PC VISION Plus library functions, and VGA graphic function as a menu driven way.