• Journal of the Korean Institute of Telematics and Electronics
• /
• v.27 no.7
• /
• pp.1025-1032
• /
• 1990

We proposed a neural network energy function for the optimal graph partitioning and its optimization method using Boltzmann Machine. We composed a Boltzmann Machine with the proposed neural network energy function, and the simulation results show that we can obtain an optimal solution with the energy function parameters of A=50, B=5, c=14 and D=10, at the Boltzmann Machine parameters of To=80 and \ulcorner0.07 for a 6-node 3-partition problem. As a result, the proposed energy function and optimization parameters are proved to be feasible for the optimal graph partitioning.

• Journal of Intelligence and Information Systems
• /
• v.29 no.3
• /
• pp.419-437
• /
• 2023

As the number of thin filers in Korea surpasses 12 million, there is a growing interest in enhancing the accuracy of assessing their credit default risk to generate additional revenue. Specifically, researchers are actively pursuing the development of default prediction models using machine learning and deep learning algorithms, in contrast to traditional statistical default prediction methods, which struggle to capture nonlinearity. Among these efforts, Graph Neural Network (GNN) architecture is noteworthy for predicting default in situations with limited data on thin filers. This is due to their ability to incorporate network information between borrowers alongside conventional credit-related data. However, prior research employing graph neural networks has faced limitations in effectively handling diverse categorical variables present in credit information. In this study, we introduce the Transformer embedded Graph Convolutional Network (TeGCN), which aims to address these limitations and enable effective default prediction for thin filers. TeGCN combines the TabTransformer, capable of extracting contextual information from categorical variables, with the Graph Convolutional Network, which captures network information between borrowers. Our TeGCN model surpasses the baseline model's performance across both the general borrower dataset and the thin filer dataset. Specially, our model performs outstanding results in thin filer default prediction. This study achieves high default prediction accuracy by a model structure tailored to characteristics of credit information containing numerous categorical variables, especially in the context of thin filers with limited data. Our study can contribute to resolving the financial exclusion issues faced by thin filers and facilitate additional revenue within the financial industry.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.8
• /
• pp.2101-2123
• /
• 2023

Recent studies have shown that the neural network-based binary code similarity detection technology performs well in vulnerability mining, plagiarism detection, and malicious code analysis. However, existing cross-architecture methods still suffer from insufficient feature characterization and low discrimination accuracy. To address these issues, this paper proposes a cross-architecture binary function similarity detection method based on composite feature model (SDCFM). Firstly, the binary function is converted into vector representation according to the proposed composite feature model, which is composed of instruction statistical features, control flow graph structural features, and application program interface calling behavioral features. Then, the composite features are embedded by the proposed hierarchical embedding network based on a graph neural network. In which, the block-level features and the function-level features are processed separately and finally fused into the embedding. In addition, to make the trained model more accurate and stable, our method utilizes the embeddings of predecessor nodes to modify the node embedding in the iterative updating process of the graph neural network. To assess the effectiveness of composite feature model, we contrast SDCFM with the state of art method on benchmark datasets. The experimental results show that SDCFM has good performance both on the area under the curve in the binary function similarity detection task and the vulnerable candidate function ranking in vulnerability search task.

• Journal of Information Processing Systems
• /
• v.19 no.6
• /
• pp.803-816
• /
• 2023

Explainable artificial intelligence is a method that explains how a complex model (e.g., a deep neural network) yields its output from a given input. Recently, graph-type data have been widely used in various fields, and diverse graph neural networks (GNNs) have been developed for graph-type data. However, methods to explain the behavior of GNNs have not been studied much, and only a limited understanding of GNNs is currently available. Therefore, in this paper, we propose an explanation method for node classification using graph convolutional networks (GCNs), which is a representative type of GNN. The proposed method finds out which features of each node have the greatest influence on the classification of that node using GCN. The proposed method identifies influential features by backtracking the layers of the GCN from the output layer to the input layer using the gradients. The experimental results on both synthetic and real datasets demonstrate that the proposed explanation method accurately identifies the features of each node that have the greatest influence on its classification.

• Steel and Composite Structures
• /
• v.46 no.6
• /
• pp.731-744
• /
• 2023

Vehicle load information is an important role in operating and ensuring the structural health of cable-stayed bridges. In this regard, an efficient and economic method is proposed for vehicle load detection based on the observed cable tension and vehicle position using a graph neural network (GNN). Datasets are first generated using the practical advanced analysis program (PAAP), a robust program for modeling and considering both geometric and material nonlinearities of bridge structures subjected to vehicle load with low computational costs. With the superiority of GNN, the proposed model is demonstrated to precisely capture complex nonlinear correlations between the input features and vehicle load in the output. Four popular machine learning methods including artificial neural network (ANN), decision tree (DT), random forest (RF), and support vector machines (SVM) are refereed in a comparison. A case study of a cable-stayed bridge with the typical truck is considered to evaluate the model's performance. The results demonstrate that the GNN-based model provides high accuracy and efficiency in prediction with satisfactory correlation coefficients, efficient determination values, and very small errors; and is a novel approach for vehicle load detection with the input data of the existing monitoring system.

• The Journal of Korea Robotics Society
• /
• v.19 no.2
• /
• pp.196-202
• /
• 2024

The ability to understand given environments and plan a sequence of actions leading to goal state is crucial for personal service robots. With recent advancements in deep learning, numerous studies have proposed methods for state representation in planning. However, previous works lack explicit information about relationships between objects when the state observation is converted to a single visual embedding containing all state information. In this paper, we introduce graph-based state representation that incorporates both object and relationship features. To leverage these advantages in addressing the task planning problem, we propose a Graph Neural Network (GNN)-based subgoal prediction model. This model can extract rich information about object and their interconnected relationships from given state graph. Moreover, a search-based algorithm is integrated with pre-trained subgoal prediction model and state transition module to explore diverse states and find proper sequence of subgoals. The proposed method is trained with synthetic task dataset collected in simulation environment, demonstrating a higher success rate with fewer additional searches compared to baseline methods.

• KIPS Transactions on Software and Data Engineering
• /
• v.8 no.5
• /
• pp.205-212
• /
• 2019

Scene graph is a kind of knowledge graph that represents both objects and their relationships found in a image. This paper proposes a 3D scene graph generation model for three-dimensional indoor environments. An 3D scene graph includes not only object types, their positions and attributes, but also three-dimensional spatial relationships between them, An 3D scene graph can be viewed as a prior knowledge base describing the given environment within that the agent will be deployed later. Therefore, 3D scene graphs can be used in many useful applications, such as visual question answering (VQA) and service robots. This proposed 3D scene graph generation model consists of four sub-networks: object detection network (ObjNet), attribute prediction network (AttNet), transfer network (TransNet), relationship prediction network (RelNet). Conducting several experiments with 3D simulated indoor environments provided by AI2-THOR, we confirmed that the proposed model shows high performance.

• Journal of the Society of Naval Architects of Korea
• /
• v.59 no.6
• /
• pp.393-399
• /
• 2022

During the ship hull design process, resistance performance estimation is generally calculated by simulation using computational fluid dynamics. Since such hull resistance performance simulation requires a lot of time and computation resources, the time taken for simulation is reduced by CPU clusters having more than tens of cores in order to complete the hull design within the required deadline of the ship owner. In this paper, we propose a method for estimating resistance performance of ship hull by simulation using a graph neural network. This method converts the 3D geometric information of the hull mesh and the physical quantity of the surface into a mathematical graph, and is implemented as a deep learning model that predicts the future simulation state from the input state. The method proposed in the resistance performance experiment of simple hull showed an average error of about 3.5 % throughout the simulation.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.537-539
• /
• 2019

구문 분석은 문장의 구조를 분석하는 자연어처리 분야로 그래프 기반 방법과 전이 기반 방법으로 나뉘어 연구되어 왔다. 본 논문에서는 그래프 기반 방식에서 높은 성능을 보이고 있는 Deep Biaffine 어텐션 모델에 별도의 High-Order 정보 추출 없이 Graph Neural Network(GNNs)을 이용하여 High-Order 정보를 학습할 수 있도록 확장한 Deep Biaffine 어텐션 GNNs을 적용하여 한국어 세종 구문 분석 셋에서 UAS : 94.44%, LAS : 92.55%의 성능을 달성하였으며 Dual Decomposition을 통해 전이 기반 한국어 구문 분석 모델과 결합하여 추가적인 성능 향상을 보였다.

• Journal of the Korean Institute of Electrical and Electronic Material Engineers
• /
• v.35 no.2
• /
• pp.134-142
• /
• 2022

As industry and technology go through advancement, it is hard to search new materials which satisfy various standards through conventional trial-and-error based research methods. Crystal Graph Convolutional Neural Network(CGCNN) is a neural network which uses material's features as train data, and predicts the material properties(formation energy, bandgap, etc.) much faster than first-principles calculation. This report introduces how to train the CGCNN model which predicts the formation energy using open database. It is anticipated that with a simple programming skill, readers could construct a model using their data and purpose. Developing machine learning model for materials science is going to help researchers who should explore large chemical and structural space to discover materials efficiently.