• Journal of the Korea Society for Simulation
• /
• v.30 no.1
• /
• pp.75-85
• /
• 2021

In this paper, we present a deep neural network-based prediction model that processes and analyzes the corporate credit and personal credit information of individual business owners as a new method to predict the default rate of individual business more accurately. In modeling research in various fields, feature selection techniques have been actively studied as a method for improving performance, especially in predictive models including many features. In this paper, after statistical verification of macroeconomic indicators (macro variables) and credit information (micro variables), which are input variables used in the default rate prediction model, additionally, through the credit information feature selection method, the final feature set that improves prediction performance was identified. The proposed credit information feature selection method as an iterative & hybrid method that combines the filter-based and wrapper-based method builds submodels, constructs subsets by extracting important variables of the maximum performance submodels, and determines the final feature set through prediction performance analysis of the subset and the subset combined set.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.6
• /
• pp.850-858
• /
• 2022

To deploy Gate Recurrent Units (GRU) on resource-constrained embedded devices, this paper presents a reconfigurable FPGA-based GRU accelerator that enables structured compression. Firstly, a dense GRU model is significantly reduced in size by hybrid quantization and structured top-k pruning. Secondly, the energy consumption on external memory access is greatly reduced by the proposed reuse computing pattern. Finally, the accelerator can handle a structured sparse model that benefits from the algorithm-hardware co-design workflows. Moreover, inference tasks can be flexibly performed using all functional dimensions, sequence length, and number of layers. Implemented on the Intel DE1-SoC FPGA, the proposed accelerator achieves 45.01 GOPs in a structured sparse GRU network without batching. Compared to the implementation of CPU and GPU, low-cost FPGA accelerator achieves 57 and 30x improvements in latency, 300 and 23.44x improvements in energy efficiency, respectively. Thus, the proposed accelerator is utilized as an early study of real-time embedded applications, demonstrating the potential for further development in the future.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.11
• /
• pp.3991-4010
• /
• 2021

The Question Similarity Measurement of Chinese Crop Diseases and Insect Pests (QSM-CCD&IP) aims to judge the user's tendency to ask questions regarding input problems. The measurement is the basis of the Agricultural Knowledge Question and Answering (Q & A) system, information retrieval, and other tasks. However, the corpus and measurement methods available in this field have some deficiencies. In addition, error propagation may occur when the word boundary features and local context information are ignored when the general method embeds sentences. Hence, these factors make the task challenging. To solve the above problems and tackle the Question Similarity Measurement task in this work, a corpus on Chinese crop diseases and insect pests(CCDIP), which contains 13 categories, was established. Then, taking the CCDIP as the research object, this study proposes a Chinese agricultural text similarity matching model, namely, the AgrCQS. This model is based on mixed information extraction. Specifically, the hybrid embedding layer can enrich character information and improve the recognition ability of the model on the word boundary. The multi-scale local information can be extracted by multi-core convolutional neural network based on multi-weight (MM-CNN). The self-attention mechanism can enhance the fusion ability of the model on global information. In this research, the performance of the AgrCQS on the CCDIP is verified, and three benchmark datasets, namely, AFQMC, LCQMC, and BQ, are used. The accuracy rates are 93.92%, 74.42%, 86.35%, and 83.05%, respectively, which are higher than that of baseline systems without using any external knowledge. Additionally, the proposed method module can be extracted separately and applied to other models, thus providing reference for related research.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.70-70
• /
• 2023

강우 발생 중 용담댐 상류로부터 용담댐으로 유입되는 유입량을 정확하게 예측하는 것은 하류 지역의 홍수 피해를 최소화하기 위한 댐의 적절한 운영에 필수적이다. 물리 기반 강우-유출 시뮬레이션 모형은 물리적 과정의 이해를 바탕으로 홍수 예측 분야에 광범위하게 사용되고 있다. 그러나 복잡한 물리 과정을 완벽히 이해하는 것은 거의 불가능하므로 다양한 가정 조건들을 이용해 복잡한 과정을 단순화하여 계산해야 하는 한계가 존재한다. 최근에는 방대한 데이터의 축적과 컴퓨터 능력의 향상으로 인해 데이터 기반 모형이 다양한 실무 문제를 해결하는 데 강력한 도구로 활용되고 있을 뿐 아니라 시뮬레이션 및 예측 등에도 다양하게 이용되고 있다. 그러나 예측 시간이 늘어날수록 입력자료로 이용되는 과거 자료와 출력자료로 이용되는 미래자료와의 상관관계가 줄어들어 모형의 성능이 저하된다. 따라서 본 연구에서는 용담댐의 시간당 유입량을 예측하기 위해 물리 기반 강우-유출 모형과 오차 보정 모형을 결합한 하이브리드 접근 방식을 제안한다. 물리 기반 강우-유출 모형으로는 HEC-HMS 모형을 사용하였으며, 오차 보정 모형에는 기계학습 모형인 인공신경망(Artificial Neural Network, ANN) 모형을 사용하였다. HEC-HMS 모형, ANN 및 하이브리드 모형(HEC-HMS + ANN)의 성능을 비교하기 위해 20 개의 홍수 사상을 모형 구축 및 검증에 사용하였다. 그 결과 하이브리드 모형은 예측 시간이 늘어날수록 HEC-HMS 및 ANN 모형보다 우수한 성능을 나타냈다. 물리모형에 기계학습을 이용한 오차 보정 절차를 통합한 경우 홍수 유출 예측의 정확성이 향상되었다. 다양한 모형의 비교 결과 본 연구에서 적용한 하이브리드 모형이 물리기반 강우-유출 모형 및 순수 기계학습 모형보다 우수한 성능을 보여줌으로써, 하이브리드 모형은 물리모형과 순수 기계학습 모형의 단점들을 보완하는데 이용할 수 있음을 나타낸다. 이 연구의 주요 목적은 강우-유출 시물레이션 모형의 오차 보정 기술에 대한 더 깊은 이해를 제공하는데 있다.

• Journal of Periodontal and Implant Science
• /
• v.53 no.1
• /
• pp.38-53
• /
• 2023

Purpose: The current Classification of Periodontal and Peri-Implant Diseases and Conditions, published and disseminated in 2018, involves some difficulties and causes diagnostic conflicts due to its criteria, especially for inexperienced clinicians. The aim of this study was to design a decision system based on machine learning algorithms by using clinical measurements and radiographic images in order to determine and facilitate the staging and grading of periodontitis. Methods: In the first part of this study, machine learning models were created using the Python programming language based on clinical data from 144 individuals who presented to the Department of Periodontology, Faculty of Dentistry, Süleyman Demirel University. In the second part, panoramic radiographic images were processed and classification was carried out with deep learning algorithms. Results: Using clinical data, the accuracy of staging with the tree algorithm reached 97.2%, while the random forest and k-nearest neighbor algorithms reached 98.6% accuracy. The best staging accuracy for processing panoramic radiographic images was provided by a hybrid network model algorithm combining the proposed ResNet50 architecture and the support vector machine algorithm. For this, the images were preprocessed, and high success was obtained, with a classification accuracy of 88.2% for staging. However, in general, it was observed that the radiographic images provided a low level of success, in terms of accuracy, for modeling the grading of periodontitis. Conclusions: The machine learning-based decision system presented herein can facilitate periodontal diagnoses despite its current limitations. Further studies are planned to optimize the algorithm and improve the results.

• Geomechanics and Engineering
• /
• v.31 no.2
• /
• pp.129-147
• /
• 2022

The properties of soil are naturally highly variable and thus, to ensure proper safety and reliability, we need to test a large number of samples across the length and depth. In pile foundations, conducting field tests are highly expensive and the traditional empirical relations too have been proven to be poor in performance. The study proposes a state-of-art Particle Swarm Optimization (PSO) hybridized Artificial Neural Network (ANN), Extreme Learning Machine (ELM) and Adaptive Neuro Fuzzy Inference System (ANFIS); and comparative analysis of metaheuristic models (ANN-PSO, ELM-PSO, ANFIS-PSO) for prediction of bearing capacity of pile foundation trained and tested on dataset of nearly 300 dynamic pile tests from the literature. A novel ensemble model of three hybrid models is constructed to combine and enhance the predictions of the individual models effectively. The authenticity of the dataset is confirmed using descriptive statistics, correlation matrix and sensitivity analysis. Ram weight and diameter of pile are found to be most influential input parameter. The comparative analysis reveals that ANFIS-PSO is the best performing model in testing phase (R² = 0.85, RMSE = 0.01) while ELM-PSO performs best in training phase (R² = 0.88, RMSE = 0.08); while the ensemble provided overall best performance based on the rank score. The performance of ANN-PSO is least satisfactory compared to the other two models. The findings were confirmed using Taylor diagram, error matrix and uncertainty analysis. Based on the results ELM-PSO and ANFIS-PSO is proposed to be used for the prediction of bearing capacity of piles and ensemble learning method of joining the outputs of individual models should be encouraged. The study possesses the potential to assist geotechnical engineers in the design phase of civil engineering projects.

• Smart Structures and Systems
• /
• v.32 no.5
• /
• pp.319-338
• /
• 2023

The study presents a new hybrid data-driven method by combining radial basis functions neural networks (RBF-NN) with the Jaya algorithm (JA) to provide effective structural health monitoring of arch dams. The novelty of this approach lies in that only one user-defined parameter is required and thus can increase its effectiveness and efficiency, as compared to other machine learning techniques that often require processing a large amount of training and testing model parameters and hyper-parameters, with high time-consuming. This approach seeks rapid damage detection in arch dams under dynamic conditions, to prevent potential disasters, by utilizing the RBF-NNN to seamlessly integrate the dynamic elastic modulus (DEM) and modal parameters (such as natural frequency and mode shape) as damage indicators. To determine the dynamic characteristics of the arch dam, the JA sequentially optimizes an objective function rooted in vibration-based data sets. Two case studies of hyperbolic concrete arch dams were carefully designed using finite element simulation to demonstrate the effectiveness of the RBF-NN model, in conjunction with the Jaya algorithm. The testing results demonstrated that the proposed methods could exhibit significant computational time-savings, while effectively detecting damage in arch dam structures with complex nonlinearities. Furthermore, despite training data contaminated with a high level of noise, the RBF-NN and JA fusion remained the robustness, with high accuracy.

• IMMUNE NETWORK
• /
• v.21 no.3
• /
• pp.23.1-23.16
• /
• 2021

Chemokines are key factors that influence the migration and maintenance of relevant immune cells into an infected tissue or a tumor microenvironment. Therefore, it is believed that the controlled administration of chemokines in the tumor microenvironment may be an effective immunotherapy against cancer. Previous studies have shown that CCL3, also known as macrophage inflammatory protein 1-alpha, facilitates the recruitment of dendritic cells (DCs) for the presentation of tumor Ags and promotes T cell activation. Here, we investigated the role of CCL3 in regulating the tumor microenvironment using a syngeneic mouse tumor model. We observed that MC38 tumors overexpressing CCL3 (CCL3-OE) showed rapid regression compared with the wild type MC38 tumors. Additionally, these CCL3-OE tumors showed an increase in the proliferative and functional tumor-infiltrating T cells. Furthermore, PD-1 immune checkpoint blockade accelerated tumor regression in the CCL3-OE tumor microenvironment. Next, we generated a modified CCL3 protein for pre-clinical use by fusing recombinant CCL3 (rCCL3) with a non-cytolytic hybrid Fc (HyFc). Administering a controlled dose of rCCL3-HyFc via subcutaneous injections near tumors was effective in tumor regression and improved survival along with activated myeloid cells and augmented T cell responses. Furthermore, combination therapy of rCCL3-HyFc with PD-1 blockade exhibited prominent effect to tumor regression. Collectively, our findings demonstrate that appropriate concentrations of CCL3 in the tumor microenvironment would be an effective adjuvant to promote anti-tumor immune responses, and suggest that administering a long-lasting form of CCL3 in combination with PD-1 blockers can have clinical applications in cancer immunotherapy.

• Annals of Hepato-Biliary-Pancreatic Surgery
• /
• v.28 no.1
• /
• pp.14-24
• /
• 2024

This study aims to assess the quality and performance of predictive models for colorectal cancer liver metastasis (CRCLM). A systematic review was performed to identify relevant studies from various databases. Studies that described or validated predictive models for CRCLM were included. The methodological quality of the predictive models was assessed. Model performance was evaluated by the reported area under the receiver operating characteristic curve (AUC). Of the 117 articles screened, seven studies comprising 14 predictive models were included. The distribution of included predictive models was as follows: radiomics (n = 3), logistic regression (n = 3), Cox regression (n = 2), nomogram (n = 3), support vector machine (SVM, n = 2), random forest (n = 2), and convolutional neural network (CNN, n = 2). Age, sex, carcinoembryonic antigen, and tumor staging (T and N stage) were the most frequently used clinicopathological predictors for CRCLM. The mean AUCs ranged from 0.697 to 0.870, with 86% of the models demonstrating clear discriminative ability (AUC > 0.70). A hybrid approach combining clinical and radiomic features with SVM provided the best performance, achieving an AUC of 0.870. The overall risk of bias was identified as high in 71% of the included studies. This review highlights the potential of predictive modeling to accurately predict the occurrence of CRCLM. Integrating clinicopathological and radiomic features with machine learning algorithms demonstrates superior predictive capabilities.

• Korean Journal of Plant Taxonomy
• /
• v.46 no.2
• /
• pp.247-255
• /
• 2016

As an attempt to determine the identity of the old trees of flowering cherries planted in the yard of the Catholic Archdiocese of Daegu, we conducted comparative phylogenetic analyses between wild and cultivated Prunus yedoensis Matsum. We generated the phylogeny (MP) and haplotype network (TCS) of 25 individuals, including wild P. yedoensis, from Jeju Island, cultivated P. ${\times}$yedoensis 'Somei-yoshino' from Korea and Japan, and P. spachiana f. ascendens (Makino) Kitam. from Jeju Island and Japan based on highly informative sequences of two cpDNA regions (rpl16 gene and trnS-trnG intergenic spacer). The wild and cultivated P. yedoensis were distinguished from each other in both the phylogeny and haplotype networks, and the old flowering cherry trees in Daegu had a cpDNA haplotype identical to that of the cultivated P. ${\times}$yedoensis 'Someiyoshino'. Compared to the cultivated P. ${\times}$yedoensis 'Somei-yoshino', wild P. yedoensis appears to have greater haplotype diversity, presumably originating from the genetic diversity of P. spachiana f. ascendens that functioned as a maternal parent in the hybrid origin of wild P. yedoensis. A future detailed study requires extensive sampling of P. spachiana f. ascendens from Japan and Korea to determine their precise phylogenetic relationships relative to wild and cultivated P. yedoensis. We concluded that the old flowering cherry trees planted in the yard of the Catholic Archdiocese of Daegu are highly likely to be of cultivated origin rather than wild types from Jeju Island, as previously speculated.