• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.12
• /
• pp.1759-1768
• /
• 2022

This paper proposes a model to detect depression-related emotions in a user's speech using wellness dialogue scripts provided by AI Hub, topic-specific daily conversation datasets, and chatbot datasets published on Github. There are 18 emotions, including depression and lethargy, in depression-related emotions, and emotion classification tasks are performed using KoBERT and KOELECTRA models that show high performance in language models. For model-specific performance comparisons, we build diverse datasets and compare classification results while adjusting batch sizes and learning rates for models that perform well. Furthermore, a person performs a multi-classification task by selecting all labels whose output values are higher than a specific threshold as the correct answer, in order to reflect feeling multiple emotions at the same time. The model with the best performance derived through this process is called the Depression model, and the model is then used to classify depression-related emotions for user utterances.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.205-208
• /
• 2022

Multi-exposure high dynamic range (HDR) image reconstruction, the task of reconstructing an HDR image from multiple low dynamic range (LDR) images in a dynamic scene, often produces ghosting artifacts caused by camera motion and moving objects and also cannot deal with washed-out regions due to over or under-exposures. While there has been many deep-learning-based methods with motion estimation to alleviate these problems, they still have limitations for severely moving scenes. They also require large parameter counts, especially in the case of state-of-the-art methods that employ attention modules. To address these issues, we propose a frequency domain approach based on the idea that the transform domain coefficients inherently involve the global information from whole image pixels to cope with large motions. Specifically we adopt Residual Fast Fourier Transform (RFFT) blocks, which allows for global interactions of pixels. Moreover, we also employ Depthwise Overparametrized convolution (DO-conv) blocks, a convolution in which each input channel is convolved with its own 2D kernel, for faster convergence and performance gains. We call this LFFNet (Lightweight Frequency Fusion Network), and experiments on the benchmarks show reduced ghosting artifacts and improved performance up to 0.6dB tonemapped PSNR compared to recent state-of-the-art methods. Our architecture also requires fewer parameters and converges faster in training.

• Steel and Composite Structures
• /
• v.45 no.2
• /
• pp.205-218
• /
• 2022

Proper calculation of splitting tensile strength (STS) of concrete has been a crucial task, due to the wide use of concrete in the construction sector. Following many recent studies that have proposed various predictive models for this aim, this study suggests and tests the functionality of three hybrid models in predicting the STS from the characteristics of the mixture components including cement compressive strength, cement tensile strength, curing age, the maximum size of the crushed stone, stone powder content, sand fine modulus, water to binder ratio, and the ratio of sand. A multi-layer perceptron (MLP) neural network incorporates invasive weed optimization (IWO), cuttlefish optimization algorithm (CFOA), and electrostatic discharge algorithm (ESDA) which are among the newest optimization techniques. A dataset from the earlier literature is used for exploring and extrapolating the STS behavior. The results acquired from several accuracy criteria demonstrated a nice learning capability for all three hybrid models viz. IWO-MLP, CFOA-MLP, and ESDA-MLP. Also in the prediction phase, the prediction products were in a promising agreement (above 88%) with experimental results. However, a comparative look revealed the ESDA-MLP as the most accurate predictor. Considering mean absolute percentage error (MAPE) index, the error of ESDA-MLP was 9.05%, while the corresponding value for IWO-MLP and CFOA-MLP was 9.17 and 13.97%, respectively. Since the combination of MLP and ESDA can be an effective tool for optimizing the concrete mixture toward a desirable STS, the last part of this study is dedicated to extracting a predictive formula from this model.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.4
• /
• pp.166-173
• /
• 2024

This study explores lip-sync detection, focusing on the synchronization between lip movements and voices in videos. Typically, lip-sync detection techniques involve cropping the facial area of a given video, utilizing the lower half of the cropped box as input for the visual encoder to extract visual features. To enhance the emphasis on the articulatory region of lips for more accurate lip-sync detection, we propose utilizing a pre-trained visual attention-based encoder. The Visual Transformer Pooling (VTP) module is employed as the visual encoder, originally designed for the lip-reading task, predicting the script based solely on visual information without audio. Our experimental results demonstrate that, despite having fewer learning parameters, our proposed method outperforms the latest model, VocaList, on the LRS2 dataset, achieving a lip-sync detection accuracy of 94.5% based on five context frames. Moreover, our approach exhibits an approximately 8% superiority over VocaList in lip-sync detection accuracy, even on an untrained dataset, Acappella.

• Journal of the Korea Computer Graphics Society
• /
• v.30 no.3
• /
• pp.51-59
• /
• 2024

Collecting a dataset with a corresponding labeled gaze vector requires a high cost in the gaze estimation field. In this paper, we suggest a data augmentation of manipulating the gaze of an original image, which improves the accuracy of the gaze estimation model when the number of given gaze labels is restricted. By conducting multi-class gaze bin classification as an auxiliary task and adjusting the latent variable of the diffusion model, the model semantically edits the gaze from the original image. We manipulate a non-binary attribute, pitch and yaw of gaze vector to a desired range and uses the edited image as an augmented train data. The improved gaze accuracy of the gaze estimation network in the semi-supervised learning validates the effectiveness of our data augmentation, especially when the number of gaze labels is 50k or less.

• Journal of Global Scholars of Marketing Science
• /
• v.19 no.3
• /
• pp.49-61
• /
• 2009

This article provides training exercises for executives into interpreting subroutine maps of executives' thinking in processing business and industrial marketing problems and opportunities. This study builds on premises that Schank proposes about learning and teaching including (1) learning occurs by experiencing and the best instruction offers learners opportunities to distill their knowledge and skills from interactive stories in the form of goal.based scenarios, team projects, and understanding stories from experts. Also, (2) telling does not lead to learning because learning requires action-training environments should emphasize active engagement with stories, cases, and projects. Each training case study includes executive exposure to decision system analysis (DSA). The training case requires the executive to write a "Briefing Report" of a DSA map. Instructions to the executive trainee in writing the briefing report include coverage in the briefing report of (1) details of the essence of the DSA map and (2) a statement of warnings and opportunities that the executive map reader interprets within the DSA map. The length maximum for a briefing report is 500 words-an arbitrary rule that works well in executive training programs. Following this introduction, section two of the article briefly summarizes relevant literature on how humans think within contexts in response to problems and opportunities. Section three illustrates the creation and interpreting of DSA maps using a training exercise in pricing a chemical product to different OEM (original equipment manufacturer) customers. Section four presents a training exercise in pricing decisions by a petroleum manufacturing firm. Section five presents a training exercise in marketing strategies by an office furniture distributer along with buying strategies by business customers. Each of the three training exercises is based on research into information processing and decision making of executives operating in marketing contexts. Section six concludes the article with suggestions for use of this training case and for developing additional training cases for honing executives' decision-making skills. Todd and Gigerenzer propose that humans use simple heuristics because they enable adaptive behavior by exploiting the structure of information in natural decision environments. "Simplicity is a virtue, rather than a curse". Bounded rationality theorists emphasize the centrality of Simon's proposition, "Human rational behavior is shaped by a scissors whose blades are the structure of the task environments and the computational capabilities of the actor". Gigerenzer's view is relevant to Simon's environmental blade and to the environmental structures in the three cases in this article, "The term environment, here, does not refer to a description of the total physical and biological environment, but only to that part important to an organism, given its needs and goals." The present article directs attention to research that combines reports on the structure of task environments with the use of adaptive toolbox heuristics of actors. The DSA mapping approach here concerns the match between strategy and an environment-the development and understanding of ecological rationality theory. Aspiration adaptation theory is central to this approach. Aspiration adaptation theory models decision making as a multi-goal problem without aggregation of the goals into a complete preference order over all decision alternatives. The three case studies in this article permit the learner to apply propositions in aspiration level rules in reaching a decision. Aspiration adaptation takes the form of a sequence of adjustment steps. An adjustment step shifts the current aspiration level to a neighboring point on an aspiration grid by a change in only one goal variable. An upward adjustment step is an increase and a downward adjustment step is a decrease of a goal variable. Creating and using aspiration adaptation levels is integral to bounded rationality theory. The present article increases understanding and expertise of both aspiration adaptation and bounded rationality theories by providing learner experiences and practice in using propositions in both theories. Practice in ranking CTSs and writing TOP gists from DSA maps serves to clarify and deepen Selten's view, "Clearly, aspiration adaptation must enter the picture as an integrated part of the search for a solution." The body of "direct research" by Mintzberg, Gladwin's ethnographic decision tree modeling, and Huff's work on mapping strategic thought are suggestions on where to look for research that considers both the structure of the environment and the computational capabilities of the actors making decisions in these environments. Such research on bounded rationality permits both further development of theory in how and why decisions are made in real life and the development of learning exercises in the use of heuristics occurring in natural environments. The exercises in the present article encourage learning skills and principles of using fast and frugal heuristics in contexts of their intended use. The exercises respond to Schank's wisdom, "In a deep sense, education isn't about knowledge or getting students to know what has happened. It is about getting them to feel what has happened. This is not easy to do. Education, as it is in schools today, is emotionless. This is a huge problem." The three cases and accompanying set of exercise questions adhere to Schank's view, "Processes are best taught by actually engaging in them, which can often mean, for mental processing, active discussion."

• Journal of Internet Computing and Services
• /
• v.21 no.1
• /
• pp.169-178
• /
• 2020

The reductions of troops/human resources, and improvement in combat power have made Korean Department of Defense actively adapt 4th Industrial Revolution technology (Artificial Intelligence, Big Data). The defense information system has been developed in various ways according to the task and the uniqueness of each military. In order to take full advantage of the 4th Industrial Revolution technology, it is necessary to improve the closed defense datamanagement system.However, the establishment and usage of data standards in all information systems for the utilization of defense big data and artificial intelligence has limitations due to security issues, business characteristics of each military, anddifficulty in standardizing large-scale systems. Based on the interworking requirements of each system, data sharing is limited through direct linkage through interoperability agreement between systems. In order to implement smart defense using the 4th Industrial Revolution technology, it is urgent to prepare a system that can share defense data and make good use of it. To technically support the defense, it is critical to develop Multi Repository Meta-Data Management (MRMM) that supports systematic standard management of defense data that manages enterprise standard and standard mapping for each system and promotes data interoperability through linkage between standards which obeys the Defense Interoperability Management Development Guidelines. We introduced MRMM, and implemented by using vocabulary similarity using machine learning and statistical approach. Based on MRMM, We expect to simplify the standardization integration of all military databases using artificial intelligence and bigdata. This will lead to huge reduction of defense budget while increasing combat power for implementing smart defense.

• Journal of Intelligence and Information Systems
• /
• v.24 no.3
• /
• pp.1-19
• /
• 2018

Large amount of data is now available for research and business sectors to extract knowledge from it. This data can be in the form of unstructured data such as audio, text, and image data and can be analyzed by deep learning methodology. Deep learning is now widely used for various estimation, classification, and prediction problems. Especially, fashion business adopts deep learning techniques for apparel recognition, apparel search and retrieval engine, and automatic product recommendation. The core model of these applications is the image classification using Convolutional Neural Networks (CNN). CNN is made up of neurons which learn parameters such as weights while inputs come through and reach outputs. CNN has layer structure which is best suited for image classification as it is comprised of convolutional layer for generating feature maps, pooling layer for reducing the dimensionality of feature maps, and fully-connected layer for classifying the extracted features. However, most of the classification models have been trained using online product image, which is taken under controlled situation such as apparel image itself or professional model wearing apparel. This image may not be an effective way to train the classification model considering the situation when one might want to classify street fashion image or walking image, which is taken in uncontrolled situation and involves people's movement and unexpected pose. Therefore, we propose to train the model with runway apparel image dataset which captures mobility. This will allow the classification model to be trained with far more variable data and enhance the adaptation with diverse query image. To achieve both convergence and generalization of the model, we apply Transfer Learning on our training network. As Transfer Learning in CNN is composed of pre-training and fine-tuning stages, we divide the training step into two. First, we pre-train our architecture with large-scale dataset, ImageNet dataset, which consists of 1.2 million images with 1000 categories including animals, plants, activities, materials, instrumentations, scenes, and foods. We use GoogLeNet for our main architecture as it has achieved great accuracy with efficiency in ImageNet Large Scale Visual Recognition Challenge (ILSVRC). Second, we fine-tune the network with our own runway image dataset. For the runway image dataset, we could not find any previously and publicly made dataset, so we collect the dataset from Google Image Search attaining 2426 images of 32 major fashion brands including Anna Molinari, Balenciaga, Balmain, Brioni, Burberry, Celine, Chanel, Chloe, Christian Dior, Cividini, Dolce and Gabbana, Emilio Pucci, Ermenegildo, Fendi, Giuliana Teso, Gucci, Issey Miyake, Kenzo, Leonard, Louis Vuitton, Marc Jacobs, Marni, Max Mara, Missoni, Moschino, Ralph Lauren, Roberto Cavalli, Sonia Rykiel, Stella McCartney, Valentino, Versace, and Yve Saint Laurent. We perform 10-folded experiments to consider the random generation of training data, and our proposed model has achieved accuracy of 67.2% on final test. Our research suggests several advantages over previous related studies as to our best knowledge, there haven't been any previous studies which trained the network for apparel image classification based on runway image dataset. We suggest the idea of training model with image capturing all the possible postures, which is denoted as mobility, by using our own runway apparel image dataset. Moreover, by applying Transfer Learning and using checkpoint and parameters provided by Tensorflow Slim, we could save time spent on training the classification model as taking 6 minutes per experiment to train the classifier. This model can be used in many business applications where the query image can be runway image, product image, or street fashion image. To be specific, runway query image can be used for mobile application service during fashion week to facilitate brand search, street style query image can be classified during fashion editorial task to classify and label the brand or style, and website query image can be processed by e-commerce multi-complex service providing item information or recommending similar item.

• Journal of Intelligence and Information Systems
• /
• v.29 no.3
• /
• pp.361-381
• /
• 2023

Research on corporate bankruptcy prediction has been focused on financial information. Since the company's financial information is updated quarterly, there is a problem that timeliness is insufficient in predicting the possibility of a company's business closure in real time. Evaluated companies that want to improve this need a method of judging the soundness of a company that uses information other than financial information to judge the soundness of a target company. To this end, as information technology has made it easier to collect non-financial information about companies, research has been conducted to apply additional variables and various methodologies other than financial information to predict corporate bankruptcy. It has become an important research task to determine whether it has an effect. In this study, we examined the impact of electronic payment-related information, which constitutes non-financial information, when predicting the closure of business operators using electronic payment service and examined the difference in closure prediction accuracy according to the combination of financial and non-financial information. Specifically, three research models consisting of a financial information model, a non-financial information model, and a combined model were designed, and the closure prediction accuracy was confirmed with six algorithms including the Multi Layer Perceptron (MLP) algorithm. The model combining financial and non-financial information showed the highest prediction accuracy, followed by the non-financial information model and the financial information model in order. As for the prediction accuracy of business closure by algorithm, XGBoost showed the highest prediction accuracy among the six algorithms. As a result of examining the relative importance of a total of 87 variables used to predict business closure, it was confirmed that more than 70% of the top 20 variables that had a significant impact on the prediction of business closure were non-financial information. Through this, it was confirmed that electronic payment-related information of non-financial information is an important variable in predicting business closure, and the possibility of using non-financial information as an alternative to financial information was also examined. Based on this study, the importance of collecting and utilizing non-financial information as information that can predict business closure is recognized, and a plan to utilize it for corporate decision-making is also proposed.