Author name cluster

Sung-Bae Cho

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

13 papers

2 author rows

AAAI Conference 2026 Conference Paper

Are Graph Transformers Necessary? Efficient Long-Range Message Passing with Fractal Nodes in MPNNs

Jeongwhan Choi
Seungjun Park
Sumin Park
Sung-Bae Cho
Noseong Park

Graph Neural Networks (GNNs) have emerged as powerful tools for learning on graph-structured data, but often struggle to balance local and global information. While graph Transformers aim to address this by enabling long-range interactions, they often overlook the inherent locality and efficiency of Message Passing Neural Networks (MPNNs). We propose a new concept called 'fractal nodes', inspired by the fractal structure observed in real-world networks. Our approach is based on the intuition that graph partitioning naturally induces fractal structure, where subgraphs often reflect the connectivity patterns of the full graph. Fractal nodes are designed to coexist with the original nodes and adaptively aggregate subgraph-level feature representations, thereby enforcing feature similarity within each subgraph. We show that fractal nodes alleviate the over-squashing problem by providing direct shortcut connections that enable long-range propagation of subgraph-level representations. Experiment results show that our method improves the expressive power of MPNNs and achieves comparable or better performance to graph Transformers while maintaining the computational efficiency of MPNN by improving the long-range dependencies of MPNN.

PDF Details DOI

AAAI Conference 2026 Conference Paper

Expandable and Differentiable Dual Memories with Orthogonal Regularization for Exemplar-free Continual Learning

Hyung-Jun Moon
Sung-Bae Cho

Continual learning methods used to force neural networks to process sequential tasks in isolation, preventing them from leveraging useful inter-task relationships and causing them to repeatedly relearn similar features or overly differentiate them. To address this problem, we propose a fully differentiable, exemplar-free expandable method composed of two complementary memories: One learns common features that can be used across all tasks, and the other combines the shared features to learn discriminative characteristics unique to each sample. Both memories are differentiable so that the network can autonomously learn latent representations for each sample. For each task, the memory adjustment module adaptively prunes critical slots and minimally expands capacity to accommodate new concepts, and orthogonal regularization enforces geometric separation between preserved and newly learned memory components to prevent interference. Experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet show that the proposed method outperforms 14 state-of-the-art methods for class-incremental learning, achieving final accuracies of 55.13%, 37.24%, and 30.11%, respectively. Additional analysis confirms that, through effective integration and utilization of knowledge, the proposed method can increase average performance across sequential tasks, and it produces feature extraction results closest to the upper bound, thus establishing a new milestone in continual learning.

PDF Details DOI

EAAI Journal 2025 Journal Article

Causally explainable artificial intelligence on deep learning model for energy demand prediction

Gatum Erlangga
Sung-Bae Cho

Accurate power demand prediction is essential for energy management in the energy sector, but it is difficult due to the factors such as greenhouse gas emissions and climate change. CNN-LSTM (Convolutional neural network-long short-term memory) neural network has demonstrated impressive performance but faces limitations in explaining its prediction results. Although XAI (explainable artificial intelligence) techniques enhance understanding through feature importance, they primarily focus on correlation rather than causality among variables in the deep learning models. To address this issue, we propose a causal XAI method for CNN-LSTM neural network with attention mechanism to predict power demand. Bayesian network is employed to provide the causal explanation with domain knowledge and relationships among observed variables and deep learning parameters (e. g. , class activation maps and attention weights). Experiments on two real datasets such as UCI (University of California, Irvine) individual household electricity dataset and REFIT (Regulatory Fitness and Performance programme of the European Commission) dataset show improvements of average 34. 84 % and 13. 63 %, respectively. It also confirms that the proposed method not only significantly outperforms state-of-the-art models in terms of prediction accuracy but also provides a causal explanation of the prediction outcome in terms of peak power usage, savings, and stability based on the observation windows, which provides actionable insight for end users to achieve power efficiency.

Details DOI

ICML Conference 2024 Conference Paper

PANDA: Expanded Width-Aware Message Passing Beyond Rewiring

Jeongwhan Choi 0002
Sumin Park
Hyowon Wi
Sung-Bae Cho
Noseong Park

Recent research in the field of graph neural network (GNN) has identified a critical issue known as "over-squashing, " resulting from the bottleneck phenomenon in graph structures, which impedes the propagation of long-range information. Prior works have proposed a variety of graph rewiring concepts that aim at optimizing the spatial or spectral properties of graphs to promote the signal propagation. However, such approaches inevitably deteriorate the original graph topology, which may lead to a distortion of information flow. To address this, we introduce an ex pand ed width- a ware ( PANDA ) message passing, a new message passing paradigm where nodes with high centrality, a potential source of over-squashing, are selectively expanded in width to encapsulate the growing influx of signals from distant nodes. Experimental results show that our method outperforms existing rewiring methods, suggesting that selectively expanding the hidden state of nodes can be a compelling alternative to graph rewiring for addressing the over-squashing.

Details

ECAI Conference 2023 Conference Paper

Adversarial Discriminator to Mitigate Gender Bias in Abusive Language Detection

Jaeil Park
Sung-Bae Cho

Abusive language detection models tend to have a gender bias problem in which the model is biased towards sentences containing identity words of specific gender groups. Previous studies to reduce bias, such as projection methods, tend to lose information in word vectors and sentence context, resulting in low detection accuracy. This paper proposes a novel method that mitigates gender bias while preserving original information by regularizing sentence embedding vectors based on information theory. Latent vectors generated by an autoencoder are debiased through dual regularization using a gender discriminator, an abuse classifier, and a decoder. While the gender discriminator labels are randomized, the discriminator confuses the gender feature, and the classifier retains the abuse information. Latent vectors are regularized through information theoretic adversarial optimization that disentangles and mitigates gender features. We show that the proposed method successfully orthogonalizes the direction of the correlated information and reduces the gender feature through calculation of subspaces and embedding vector visualization. Moreover, the proposed method maintains the highest accuracy among the four state-of-the-art bias mitigation methods and shows superior performance in reducing gender bias in four different Twitter datasets for abusive language detection.

Details

ICML Conference 2023 Conference Paper

GREAD: Graph Neural Reaction-Diffusion Networks

Jeongwhan Choi 0002
Seoyoung Hong 0001
Noseong Park
Sung-Bae Cho

Graph neural networks (GNNs) are one of the most popular research topics for deep learning. GNN methods typically have been designed on top of the graph signal processing theory. In particular, diffusion equations have been widely used for designing the core processing layer of GNNs, and therefore they are inevitably vulnerable to the notorious oversmoothing problem. Recently, a couple of papers paid attention to reaction equations in conjunctions with diffusion equations. However, they all consider limited forms of reaction equations. To this end, we present a reaction-diffusion equation-based GNN method that considers all popular types of reaction equations in addition to one special reaction equation designed by us. To our knowledge, our paper is one of the most comprehensive studies on reaction-diffusion equation-based GNNs. In our experiments with 9 datasets and 28 baselines, our method, called GREAD, outperforms them in a majority of cases. Further synthetic data experiments show that it mitigates the oversmoothing problem and works well for various homophily rates.

Details

EAAI Journal 2019 Journal Article

An ensemble semi-supervised learning method for predicting defaults in social lending

Aleum Kim
Sung-Bae Cho

Social lending is made between peers, and with the risk that the investor can take direct damages from the borrower’s failure to repay, accurate default prediction for borrowers is important. The repayment result can be known after the end of the repayment period, and such data is limited. However, social loans are matched online in real time and large amounts of unlabeled data are being generated. In this paper, we propose a method to combine label propagation and transductive support vector machine (TSVM) with Dempster–Shafer theory for accurate default prediction of social lending using unlabeled data. In order to train a lot of data effectively, we ensemble semi-supervised learning methods with different characteristics. Label propagation is performed so that data having similar features are assigned to the same class and TSVM makes moving away data having different features. Dempster–Shafer fusion method allows accurate labeling by exploiting the merits of the two methods. Experiments are performed using the open data set from Lending Club. The accuracy of the proposed method is improved by about 10% against that of the model using only labeled data, and more accurate labeling can be performed through the proposed ensemble method.

Details DOI

EAAI Journal 2013 Journal Article

DE+RBFNs based classification: A special attention to removal of inconsistency and irrelevant features

Ch. Sanjeev Kumar Dash
Aditya Prakash Dash
Satchidananda Dehuri
Sung-Bae Cho
Gi-Nam Wang

A novel approach for the classification of both balanced and imbalanced dataset is developed in this paper by integrating the best attributes of radial basis function networks and differential evolution. In addition, a special attention is given to handle the problem of inconsistency and removal of irrelevant features. Removing data inconsistency and inputting optimal and relevant set of features to a radial basis function network may greatly enhance the network efficiency (in terms of accuracy), at the same time compact its size. We use Bayesian statistics for making the dataset consistent, information gain theory (a kind of filter approach) for reducing the features, and differential evolution for tuning center, spread and bias of radial basis function networks. The proposed approach is validated with a few benchmarked highly skewed and balanced dataset retrieved from University of California, Irvine (UCI) repository. Our experimental result demonstrates promising classification accuracy, when data inconsistency and feature selection are considered to design this classifier.

Details DOI

AIIM Journal 2006 Journal Article

The classification of cancer based on DNA microarray data that uses diverse ensemble genetic programming

Jin-Hyuk Hong
Sung-Bae Cho

Object The classification of cancer based on gene expression data is one of the most important procedures in bioinformatics. In order to obtain highly accurate results, ensemble approaches have been applied when classifying DNA microarray data. Diversity is very important in these ensemble approaches, but it is difficult to apply conventional diversity measures when there are only a few training samples available. Key issues that need to be addressed under such circumstances are the development of a new ensemble approach that can enhance the successful classification of these datasets. Materials and methods An effective ensemble approach that does use diversity in genetic programming is proposed. This diversity is measured by comparing the structure of the classification rules instead of output-based diversity estimating. Results Experiments performed on common gene expression datasets (such as lymphoma cancer dataset, lung cancer dataset and ovarian cancer dataset) demonstrate the performance of the proposed method in relation to the conventional approaches. Conclusion Diversity measured by comparing the structure of the classification rules obtained by genetic programming is useful to improve the performance of the ensemble classifier.

Details DOI

ICRA Conference 2005 Conference Paper

A Hierarchical Bayesian Network for Mixed-Initiative Human-Robot Interaction

Jin-Hyuk Hong
Youn-Suk Song
Sung-Bae Cho

The service robot supports people in their daily activities, while the interaction between humans and robots seems to be an important part of its performance. Dialogue may be beneficial to the robot to increase the flexibility and facility of the interaction. Traditional robots have merely dealt with simple queries like commands, but in conversation people often omit some words because of the background knowledge or the context of the conversation. Since environments contain various uncertainties, managing the context of a dialogue or the uncertainties should be necessary to support smarter service robots. In order to establish a natural communication between people and robots, we have been investigating the use of mixed-initiative interaction that prompts for missing concepts and clarifies for spurious concepts. Hierarchically designed Bayesian networks are presented for the mixed-initiative interaction. A simulation and a real robot are constructed for the demonstration of the proposed method, and experiments also show the usefulness.

Details

EAAI Journal 2003 Journal Article

Extracting intuitive strokes in complex structured patterns with domain knowledge

Sung-Bae Cho
Dong-Hyeop Han

A standard iterative thinning algorithm which has been widely used to extract features for character recognition may destroy information due to several defects such as spurious loops in the skeleton and deformation in touched strokes. This is because most thinning algorithms rely on the steady erosion of character boundaries while maintaining the connectivity of the shape. To solve this problem, this paper proposes a knowledge-based thinning method which removes the spurious loops by a preprocessing stage and makes use of average stroke widths and domain knowledge on Hangul (Korean script) to extract intuitive strokes. The experimental results on 2000 handwritten Hangul characters in PE92 benchmark database indicate that the proposed method has reduced the number of defects and led to more intuitive strokes.

Details DOI

EAAI Journal 2000 Journal Article

Application of interactive genetic algorithm to fashion design

Hee-Su Kim
Sung-Bae Cho

In general, computer-aided design support systems have got an approach of traditional artificial intelligence, which statistically analyzes data such as the behavior of designer, to extract formal design behavior. This approach, however, can neither deal with continuous change of fashion nor reflect personal taste well, as it just depends on large amount of collected data. To overcome this sort of problem interactive genetic algorithm (IGA) has been recently proposed, as a new trend of evolutionary computation. IGA uses human's response as fitness value when the fitness function cannot be explicitly defined. This enables IGA to be applied to artistic domains, and we propose a fashion design aid system using it. Unlike the previous works that attempt to model the dress design by several spline curves, the proposed system is based on a new encoding scheme that practically describes a dress with three parts: body and neck, sleeve, and skirt. By incorporating the domain-specific knowledge into the genotype, we could develop a more realistic design aid system for women’s dress. We have implemented the system with OpenGL and VRML to enhance the system interface. The experiments with several human subjects show that the IGA approach to dress design aid system is promising.

Details DOI

EAAI Journal 2000 Journal Article

Automated vectorization of cartographic maps by a knowledge-based system

Kyong-Ho Lee
Sung-Bae Cho
Yoon-Chul Choy

Developing an automated vectorizing system as an input method for a geographic information system (GIS) is of extreme importance due to the fact that an input process takes a lot of time and cost in constructing a GIS. Most vectorizing systems require users to set the parameters as appropriately as possible for a particular map image, but it is quite difficult for a novice to adjust the parameters appropriately. This paper proposes a knowledge-based system for automated vectorization, allowing an appropriate choice of the parameters. Since thinning of the input image to produce a skeleton of unit width is a prerequisite for the automated vectorization among several steps, the performance of representative thinning algorithms is systematically evaluated in various map images, and appropriate rules for the maps are devised. Each rule in the knowledge base is characterized by the type of map, and by the resolution, line width, slope and protrusions. Experimental results with various map images show that the proposed system is superior in terms of performance and convenience of use.

Details DOI