Jongho Im
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Jongho Im has a total of 61 co-authors. Here are the top co-authors by number of shared publications:
- Nathaniel Kang
Yonsei University Seoul Korea
5 shared publications Hae-Hwan Lee
4 shared publicationsSeunghwan Park
4 shared publications- Jewoo Kim
Department of Apparel, Events, and Hospitality Management, Iowa State University, Ames, IA, USA
3 shared publications - Joon Yeon Choeh
Department of Software, Sejong University, Seoul, Republic of Korea
3 shared publications - Young Min Kim
Department of Statistics, Kyungpook National University
3 shared publications
Latest Publications
- 10 November 2026
- Published by Wiley in WIREs Computational Statistics
- Vol. 18 (4) , e70078
A Survey on Tabular Data Synthesis: Generation, Evaluation, and Benchmarking
Tabular data synthesis has attracted growing attention as a means of addressing data scarcity, supporting data sharing, and augmenting datasets for analysis. Recent advances in generative modeling motivate an updated review of how emerging synthesis methods relate to established statistical approaches and how their outputs should be evaluated. This review brings together classical statistical methods, deep generative models, and recent approaches involving large language models (LLMs). We examine their generation mechanisms, modeling assumptions, and limitations, and discuss evaluation from three complementary perspectives: distributional similarity, downstream utility, and privacy. An illustrative benchmark of twelve generator configurations on six datasets complements the review by examining dataset‐ and metric‐dependent behavior under the reported experimental protocol. The comparison highlights the need to interpret similarity and utility scores alongside privacy‐related diagnostics and computational requirements. We also discuss challenges involving mixed data types, rare observations, reproducibility, and responsible deployment. The review identifies open questions about how synthetic data should be generated and validated for particular uses, including when LLMs are useful as direct generators or auxiliary components.
- 10 November 2026
- Published by Wiley in WIREs Computational Statistics
- Vol. 18 (4) , e70078
A Survey on Tabular Data Synthesis: Generation, Evaluation, and Benchmarking
Tabular data synthesis has attracted growing attention as a means of addressing data scarcity, supporting data sharing, and augmenting datasets for analysis. Recent advances in generative modeling motivate an updated review of how emerging synthesis methods relate to established statistical approaches and how their outputs should be evaluated. This review brings together classical statistical methods, deep generative models, and recent approaches involving large language models (LLMs). We examine their generation mechanisms, modeling assumptions, and limitations, and discuss evaluation from three complementary perspectives: distributional similarity, downstream utility, and privacy. An illustrative benchmark of twelve generator configurations on six datasets complements the review by examining dataset‐ and metric‐dependent behavior under the reported experimental protocol. The comparison highlights the need to interpret similarity and utility scores alongside privacy‐related diagnostics and computational requirements. We also discuss challenges involving mixed data types, rare observations, reproducibility, and responsible deployment. The review identifies open questions about how synthetic data should be generated and validated for particular uses, including when LLMs are useful as direct generators or auxiliary components.
- Open Access
- 4 June 2026
- Published by Elsevier in SSRN Electronic Journal
Improving Temporal Consistency in Text-Based Indices with Rank-Informed Calibration
Text-based indices from news and social media track latent constructs such as risk, sentiment, and uncertainty, but evolving language and coverage induce temporal measurement error that undermines longitudinal comparability.We cast calibration as a rank-constrained convex program: with partial ordinal information on a subset of time points, a convex monotone M-estimator yields a unique rank-consistent fit; the squared-loss specialization reduces to pool-adjacent-violators estimation with standard isotonic guarantees.Stochastic residual augmentation preserves expected ordering while restoring variability, and a regularized supervised model propagates the calibrated scale to the full series without additional rank labels.A semiconductor risk case study with expert anchors shows strong leave-one-out generalization, alignment with independent volatility proxies, and improved temporal coherence relative to distribution matching and regression-style comparators.
- 4 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessImproving Temporal Consistency in Text-Based Indices with Rank-Informed Calibration
Text-based indices from news and social media track latent constructs such as risk, sentiment, and uncertainty, but evolving language and coverage induce temporal measurement error that undermines longitudinal comparability.We cast calibration as a rank-constrained convex program: with partial ordinal information on a subset of time points, a convex monotone M-estimator yields a unique rank-consistent fit; the squared-loss specialization reduces to pool-adjacent-violators estimation with standard isotonic guarantees.Stochastic residual augmentation preserves expected ordering while restoring variability, and a regularized supervised model propagates the calibrated scale to the full series without additional rank labels.A semiconductor risk case study with expert anchors shows strong leave-one-out generalization, alignment with independent volatility proxies, and improved temporal coherence relative to distribution matching and regression-style comparators.
- Open Access
- 27 May 2026
- Published by Elsevier in SSRN Electronic Journal
Not All Large Language Model Deployments Are Created Equal: A Taxonomy-Driven Survey of Security, Defense, and Governance
The rapid enterprise adoption of Large Language Models (LLMs) has generated an expanding attack surface that existing surveys address monolithically, leaving engineering practitioners without deployment-specific guidance. This survey organizes the LLM security and governance landscape along two system design dimensions—user accountability boundary and data exposure temporality—yielding four deployment archetypes: Secured Repository, Managed Pipeline, Fortified Service, and Contested Interface. Ten attack families and twelve defense categories are mapped onto these archetypes with quantitative security-utility tradeoff characterizations, enabling context-specific threat prioritization for real-world applications. A comparison of five governance frameworks reveals critical coverage gaps, particularly for continuous-exposure deployments, and identifies governance evolution pathways organizations traverse as system architectures mature. By connecting deployment context to threat relevance, defense effectiveness, and governance applicability, this survey provides actionable engineering guidance for designing and securing LLM-powered software systems.
- 27 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessNot All Large Language Model Deployments Are Created Equal: A Taxonomy-Driven Survey of Security, Defense, and Governance
The rapid enterprise adoption of Large Language Models (LLMs) has generated an expanding attack surface that existing surveys address monolithically, leaving engineering practitioners without deployment-specific guidance. This survey organizes the LLM security and governance landscape along two system design dimensions—user accountability boundary and data exposure temporality—yielding four deployment archetypes: Secured Repository, Managed Pipeline, Fortified Service, and Contested Interface. Ten attack families and twelve defense categories are mapped onto these archetypes with quantitative security-utility tradeoff characterizations, enabling context-specific threat prioritization for real-world applications. A comparison of five governance frameworks reveals critical coverage gaps, particularly for continuous-exposure deployments, and identifies governance evolution pathways organizations traverse as system architectures mature. By connecting deployment context to threat relevance, defense effectiveness, and governance applicability, this survey provides actionable engineering guidance for designing and securing LLM-powered software systems.
- 27 May 2026
- Published by Elsevier in Tourism Management
- Vol. 117
Can we use generative AI for tourism research? A guide to applying and validating large language models for text analytics
- 27 May 2026
- Published by Elsevier in Tourism Management
- Vol. 117
Can we use generative AI for tourism research? A guide to applying and validating large language models for text analytics
- Open Access
- 26 May 2026
- Published by Elsevier in SSRN Electronic Journal
EAME: Extreme-Aware Mixture-of-Experts for Electricity Forecasting
In electricity system time-series forecasting, accurately predicting extreme values is essential for system planning, resource allocation, and operational reliability.Despite its practical importance, extreme-value forecasting remains challenging.Prior approaches have addressed this problem by modifying the training objective or by adopting Mixture-of-Experts (MoE) frameworks.However, these methods often sacrifice accuracy on normal samples or rely on explicit data partitioning to handle extreme values.To overcome these limitations, we propose EAME, an extreme-aware MoE-based method that models general and extreme regimes without explicit data partitioning. EAME consists of a general pattern expert that focuses on dominant temporal patterns, an extreme pattern expert that focuses on extreme patterns using the proposed Asymmetric Extreme Loss, and a combination expert that dynamically determines their contributions. The final prediction is obtained through a weighted aggregation of expert outputs.Extensive experiments on real-world electricity datasets demonstrate that EAME consistently improves forecasting performance, reducing RMSE by 1.5\%–44.1\% and Extreme RMSE (RMSE computed on extreme samples) by 9.1\%–53.3\% relative to baseline methods.Furthermore, we introduce a blackout cost metric to measure the economic losses associated with extreme scenarios, and conduct feature importance analysis to improve the interpretability of EAME.
- 26 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessEAME: Extreme-Aware Mixture-of-Experts for Electricity Forecasting
In electricity system time-series forecasting, accurately predicting extreme values is essential for system planning, resource allocation, and operational reliability.Despite its practical importance, extreme-value forecasting remains challenging.Prior approaches have addressed this problem by modifying the training objective or by adopting Mixture-of-Experts (MoE) frameworks.However, these methods often sacrifice accuracy on normal samples or rely on explicit data partitioning to handle extreme values.To overcome these limitations, we propose EAME, an extreme-aware MoE-based method that models general and extreme regimes without explicit data partitioning. EAME consists of a general pattern expert that focuses on dominant temporal patterns, an extreme pattern expert that focuses on extreme patterns using the proposed Asymmetric Extreme Loss, and a combination expert that dynamically determines their contributions. The final prediction is obtained through a weighted aggregation of expert outputs.Extensive experiments on real-world electricity datasets demonstrate that EAME consistently improves forecasting performance, reducing RMSE by 1.5\%–44.1\% and Extreme RMSE (RMSE computed on extreme samples) by 9.1\%–53.3\% relative to baseline methods.Furthermore, we introduce a blackout cost metric to measure the economic losses associated with extreme scenarios, and conduct feature importance analysis to improve the interpretability of EAME.
- 21 October 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 186
Generalized additive models for mixed-data regression using informal data
- 21 October 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 186
Generalized additive models for mixed-data regression using informal data
- 15 October 2025
- Published by Elsevier in Pattern Recognition
- Vol. 172
Calibrated mixup for imbalanced regression on tabular data
- 15 October 2025
- Published by Elsevier in Pattern Recognition
- Vol. 172
Calibrated mixup for imbalanced regression on tabular data
- 8 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (12) , 19065-19089
Optimizing Federated Learning: Addressing Key Challenges in Real-World Applications
Federated learning (FL) has emerged as a powerful solution for training machine learning models across decentralized data sources, preserving privacy by keeping data on local devices. However, the method introduces significant challenges, particularly in managing asymmetric client availability, addressing statistical and system heterogeneity, and maintaining privacy while optimizing model performance. In this study, we critically examine these challenges and present a systematic taxonomy that categorizes FL research into key areas: optimization algorithms, client selection strategies, heterogeneity handling, and privacy preservation. We analyze how balancing these aspects is essential for the successful deployment of FL in real-world applications. Additionally, we explore future research directions and provide practical insights into how FL can be optimized and applied in industries such as healthcare, energy, and autonomous systems. This survey offers a comprehensive resource for researchers and practitioners seeking to enhance the robustness, fairness, and efficiency of FL systems in diverse and dynamic environments.
- 8 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (12) , 19065-19089
Optimizing Federated Learning: Addressing Key Challenges in Real-World Applications
Federated learning (FL) has emerged as a powerful solution for training machine learning models across decentralized data sources, preserving privacy by keeping data on local devices. However, the method introduces significant challenges, particularly in managing asymmetric client availability, addressing statistical and system heterogeneity, and maintaining privacy while optimizing model performance. In this study, we critically examine these challenges and present a systematic taxonomy that categorizes FL research into key areas: optimization algorithms, client selection strategies, heterogeneity handling, and privacy preservation. We analyze how balancing these aspects is essential for the successful deployment of FL in real-world applications. Additionally, we explore future research directions and provide practical insights into how FL can be optimized and applied in industries such as healthcare, energy, and autonomous systems. This survey offers a comprehensive resource for researchers and practitioners seeking to enhance the robustness, fairness, and efficiency of FL systems in diverse and dynamic environments.
- Open Access
- 4 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 11 (3) , 1464-1474
Online News-Based Economic Sentiment Index
The accurate prediction of industry trends has become increasingly challenging because of unforeseen events. To address this challenge, this study proposes a deep learning approach to generate an economic sentiment index by integrating Natural Language Processing (NLP) models and image-clustering techniques. We first employ sampling techniques to create standardized online news datasets. Feature engineering techniques from the Korean Bidirectional Encoder Representations from Transformers (KoBERT) model are then used to generate relevance and sentiment scores for the textual data. Further, to enhance visualization and clustering, we transform the textual data into joint plot images, which are grouped into distinct clusters based on news categories. Finally, using Multi-criteria Decision Analysis, the various scores and cluster information are synthesized to generate the final economic sentiment index. This approach improves visualization and enhances the interpretability of the generated index. The proposed algorithm is applied to construct a new economic sentiment index for the Information and Communications Technology (ICT) industry in South Korea.
- 4 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Big Data
- Vol. 11 (3) , 1464-1474
Open AccessOnline News-Based Economic Sentiment Index
The accurate prediction of industry trends has become increasingly challenging because of unforeseen events. To address this challenge, this study proposes a deep learning approach to generate an economic sentiment index by integrating Natural Language Processing (NLP) models and image-clustering techniques. We first employ sampling techniques to create standardized online news datasets. Feature engineering techniques from the Korean Bidirectional Encoder Representations from Transformers (KoBERT) model are then used to generate relevance and sentiment scores for the textual data. Further, to enhance visualization and clustering, we transform the textual data into joint plot images, which are grouped into distinct clusters based on news categories. Finally, using Multi-criteria Decision Analysis, the various scores and cluster information are synthesized to generate the final economic sentiment index. This approach improves visualization and enhances the interpretability of the generated index. The proposed algorithm is applied to construct a new economic sentiment index for the Information and Communications Technology (ICT) industry in South Korea.