• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.12
• /
• pp.1769-1776
• /
• 2022

Automatically summarizing long sentences is an important technique. The BART model is one of the widely used models in the summarization task. In general, in order to generate a summarization model of a specific domain, fine-tuning is performed by re-training a language model trained on a large dataset to fit the domain. The fine-tuning is usually done by changing the number of nodes in the last fully connected layer. However, in this paper, we propose a fine-tuning method by adding an attention layer, which has been recently applied to various models and shows good performance. In order to evaluate the performance of the proposed method, various experiments were conducted, such as accumulating layers deeper, fine-tuning without skip connections during the fine tuning process, and so on. As a result, the BART model using two attention layers with skip connection shows the best score.

• Convergence Security Journal
• /
• v.22 no.4
• /
• pp.17-24
• /
• 2022

Recently, machine learning-based cyber attack detection and classification research has been actively conducted, achieving a high level of detection accuracy. However, low-spec IoT devices and large-scale network traffic make it difficult to apply machine learning-based detection models in IoT environment. Therefore, In this paper, we propose an efficient IoT attack detection and classification method through PCA(Principal Component Analysis) and LightGBM(Light Gradient Boosting Model) using datasets collected in a MQTT(Message Queuing Telementry Transport) IoT protocol environment that is also used in the defense field. As a result of the experiment, even though the original dataset was reduced to about 15%, the performance was almost similar to that of the original. It also showed the best performance in comparative evaluation with the four dimensional reduction techniques selected in this paper.

• Computers and Concrete
• /
• v.32 no.2
• /
• pp.149-163
• /
• 2023

One of the main design parameters traditionally utilized in projects of geotechnical engineering is the uniaxial compressive strength. The present paper employed three artificial intelligence methods, i.e., the stochastic fractal search (SFS), the multi-verse optimization (MVO), and the vortex search algorithm (VSA), in order to determine the compressive strength of concrete (CSC). For the same reason, 1030 concrete specimens were subjected to compressive strength tests. According to the obtained laboratory results, the fly ash, cement, water, slag, coarse aggregates, fine aggregates, and SP were subjected to tests as the input parameters of the model in order to decide the optimum input configuration for the estimation of the compressive strength. The performance was evaluated by employing three criteria, i.e., the root mean square error (RMSE), mean absolute error (MAE), and the determination coefficient (R²). The evaluation of the error criteria and the determination coefficient obtained from the above three techniques indicates that the SFS-MLP technique outperformed the MVO-MLP and VSA-MLP methods. The developed artificial neural network models exhibit higher amounts of errors and lower correlation coefficients in comparison with other models. Nonetheless, the use of the stochastic fractal search algorithm has resulted in considerable enhancement in precision and accuracy of the evaluations conducted through the artificial neural network and has enhanced its performance. According to the results, the utilized SFS-MLP technique showed a better performance in the estimation of the compressive strength of concrete (R²=0.99932 and 0.99942, and RMSE=0.32611 and 0.24922). The novelty of our study is the use of a large dataset composed of 1030 entries and optimization of the learning scheme of the neural prediction model via a data distribution of a 20:80 testing-to-training ratio.

• Korean Journal of Cognitive Science
• /
• v.34 no.3
• /
• pp.197-226
• /
• 2023

Generative Text Summarization is one of the Natural Language Processing tasks. It generates a short abbreviated summary while preserving the content of the long text. ROUGE is a widely used lexical-overlap based metric for text summarization models in generative summarization benchmarks. Although it shows very high performance, the studies report that 30% of the generated summary and the text are still inconsistent. This paper proposes a methodology for evaluating the performance of the summary model without using the correct summary. AggreFACT is a human-annotated dataset that classifies the types of errors in neural text summarization models. Among all the test candidates, the two cases, generation summary, and when errors occurred throughout the summary showed the highest correlation results. We observed that the proposed evaluation score showed a high correlation with models finetuned with BART and PEGASUS, which is pretrained with a large-scale Transformer structure.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.18 no.5
• /
• pp.999-1008
• /
• 2023

Early prediction of chronic diseases such as diabetes is an important issue, and improving the accuracy of diabetes prediction is especially important. Various machine learning and deep learning-based methodologies are being introduced for diabetes prediction, but these technologies require large amounts of data for better performance than other methodologies, and the learning cost is high due to complex data models. In this study, we aim to verify the claim that DNN using the pima dataset and k-fold cross-validation reduces the efficiency of diabetes diagnosis models. Machine learning classification methods such as decision trees, SVM, random forests, logistic regression, KNN, and various ensemble techniques were used to determine which algorithm produces the best prediction results. After training and testing all classification models, the proposed system provided the best results on XGBoost classifier with ADASYN method, with accuracy of 81%, F1 coefficient of 0.81, and AUC of 0.84. Additionally, a domain adaptation method was implemented to demonstrate the versatility of the proposed system. An explainable AI approach using the LIME and SHAP frameworks was implemented to understand how the model predicts the final outcome.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.2
• /
• pp.79-86
• /
• 2024

In this paper, a novel visual tracking method which can utilize the context of object regions is presented. Conventional methods have the inherent problem of treating all candidate regions independently, where the tracker could not successfully discriminate regions with similar appearances. This was due to lack of contextual modeling in a given scene, where all candidate object regions should be taken into consideration when choosing a single region. The goal of the proposed method is to encourage feature exchange between candidate regions to improve the discriminability between similar regions. It improves upon conventional methods that only consider a single region, and is implemented by employing the MLP-Mixer model for enhanced feature exchange between regions. By implementing channel-wise, inter-region interaction operation between candidate features, contextual information of regions can be embedded into the individual feature representations. To evaluate the performance of the proposed tracker, the large-scale LaSOT dataset is used, and the experimental results show a competitive AUC performance of 0.560 while running at a real-time speed of 65 fps.

• Journal of Yeungnam Medical Science
• /
• v.40 no.4
• /
• pp.402-411
• /
• 2023

Background: We hypothesized that fatty liver disease (FLD) is associated with a high prevalence of hearing loss (HL) owing to metabolic disturbances. This study aimed to evaluate the association between FLD and HL in a large sample of the Korean population. Methods: We used a dataset of adults who underwent routine voluntary health checkups (n=21,316). Fatty liver index (FLI) was calculated using Bedogni's equation. The patients were divided into two groups: the non-FLD (NFLD) group (n=18,518, FLI <60) and the FLD group (n=2,798, FLI ≥60). Hearing thresholds were measured using an automatic audiometer. The average hearing threshold (AHT) was calculated as the pure-tone average at four frequencies (0.5, 1, 2, and 3 kHz). HL was defined as an AHT of >40 dB. Results: HL was observed in 1,370 (7.4%) and 238 patients (8.5%) in the NFLD and FLD groups, respectively (p=0.041). Compared with the NFLD group, the odds ratio for HL in the FLD group was 1.16 (p=0.040) and 1.46 (p<0.001) in univariate and multivariate logistic regression analyses, respectively. Linear regression analyses revealed that FLI was positively associated with AHT in both univariate and multivariate analyses. Analyses using a propensity score-matched cohort showed trends similar to those using the total cohort. Conclusion: FLD and FLI were associated with poor hearing thresholds and HL. Therefore, active monitoring of hearing impairment in patients with FLD may be helpful for early diagnosis and treatment of HL in the general population.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.22 no.6
• /
• pp.157-167
• /
• 2023

The use of big data for transportation often involves using data that includes personal information, such as the driver's driving routes and coordinates. This study explores the creation of a route choice prediction model using a large dataset from mobile navigation apps using federated learning. This privacy-focused method used distributed computing and individual device usage. This study established preprocessing and analysis methods for driver data that can be used in route choice modeling and compared the performance and characteristics of widely used learning methods with federated learning methods. The performance of the model through federated learning did not show significantly superior results compared to previous models, but there was no substantial difference in the prediction accuracy. In conclusion, federated learning-based prediction models can be utilized appropriately in areas sensitive to privacy without requiring relatively high predictive accuracy, such as a driver's preferred route choice.

• ALGAE
• /
• v.39 no.2
• /
• pp.97-108
• /
• 2024

The Ralfsiaceae family, part of the Ralfsiales order and consisting of crustose brown algae, includes five genera: Analipus, Endoplura, Fissipedicella, Heteroralfsia, and Ralfsia. In this study, a novel crustose genus named Ramipedicella gen. nov. is introduced within the Ralfsiaceae based on molecular and morphological analyses. Phylogenetic analyses using both concatenated dataset (rbcL + COI-5P genes) and rbcL indicate that the crustose brown algae that we collected from Korea and Russia form a unique grouping within the Ralfsiaceae. This grouping is strongly supported by both bootstrap analysis and Bayesian posterior probabilities. The genetic differences in the rbcL and COI-5P sequences between Ramipedicella and other genera within Ralfsiaceae range from 6.7 to 9.3% for rbcL and from 15.5 to 20.8% for COI-5P. Ramipedicella is characterized by crustose thalli having new crusts growing on top of old ones with a hypothallial basal layer and erect perithallial filaments, long cells with width-to-length ratio of 1 : 1-16, single chloroplast per cell, plurangia with one to several sterile cells, one to several unangia produced from unicellular stalks or from the lateral-basal region to the paraphyses, and unangia arising sequencially in irregularly branched specialized filaments. Ramipedicella, the recently identified genus, comprises two distinct species. Ramipedicella miniloba, the type species, is distinguished by crusts with small lobes, numerous hair tufts, plurangia terminated by 1-4 sterile cells, and large oblong unangia. Ramipedicella longicellularis is identified by generally smooth crusts, absence of phaeophycean hairs, plurangia terminated by 1-2 apical sterile cells, and smaller mostly oblanceolate unangia.

• Journal of information and communication convergence engineering
• /
• v.22 no.1
• /
• pp.33-43
• /
• 2024

Automated text summarization (ATS) systems rely on language resources as datasets. However, creating these datasets is a complex and labor-intensive task requiring linguists to extensively annotate the data. Consequently, certain public datasets for ATS, particularly in languages such as Thai, are not as readily available as those for the more popular languages. The primary objective of the ATS approach is to condense large volumes of text into shorter summaries, thereby reducing the time required to extract information from extensive textual data. Owing to the challenges involved in preparing language resources, publicly accessible datasets for Thai ATS are relatively scarce compared to those for widely used languages. The goal is to produce concise summaries and accelerate the information extraction process using vast amounts of textual input. This study introduced ThEconSum, an ATS architecture specifically designed for Thai language, using economy-related data. An evaluation of this research revealed the significant remaining tasks and limitations of the Thai language.