Papers by Jaehyung Kim
Few-shot Personalization of LLMs with Mis-aligned Responses (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing approaches have limited successes in personalizing large language models due to the lack of personalized learning or the reliance on shared personal data. |
| Approach: | They propose a few-shot personalization of large language models with mis-aligned responses using LLMs by learning a set of personalized prompts for each user based on user profile and examples of previous opinions. |
| Outcome: | The proposed method significantly improves performance across benchmarks compared to best-performing baselines. |
LimaCost: Data Valuation for Instruction Tuning of Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Instruction tuning is an effective approach for aligning large language models with human intentions. |
| Approach: | They propose a data quality measure that exhibits a strong correlation with model performance. |
| Outcome: | The proposed measure exhibits a strong correlation with model performance. |
Revisiting the Uniform Information Density Hypothesis in LLM Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | a recent study has highlighted the fragility of Chain-of-Thought reasoning . a hypothesis suggests that effective communication is achieved by maintaining a stable flow of information. |
| Approach: | They propose a framework to quantify uniformity of information flow at local and global levels . they propose entropy-based stepwise density metric to quantify this phenomenon . |
| Outcome: | The proposed framework outperforms alternative signals as predictors of reasoning quality. |
Learning to Correct for QA Reasoning with Black-box LLMs (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to improve reasoning capability of large language models rely on accessibility or require significantly increased train- and inference-time costs. |
| Approach: | They propose a method to improve QA reasoning of large language models in a black-box setting by using a trained adaptation model to perform a seq2seq mapping from the often-imperfect reasonings of the original LLM to the correct or improved reasonings. |
| Outcome: | The proposed approach significantly improves reasoning accuracy across various QA benchmarks compared to the best-performing adaptation baselines. |
KoLEG: On-the-Fly Korean Legal Knowledge Editing with Continuous Retrieval (2025.findings-emnlp)
Copied to clipboard
Jaehyung Seo, Dahyun Jung, Jaewook Lee, Yongchan Chun, Dongjun Kim, Hwijung Ryu, Donghoon Shin, Heuiseok Lim
| Challenge: | a recent study shows that Korean legal knowledge is subject to frequent temporal updates driven by societal needs and government policies. |
| Approach: | They propose a Korean Legal knowledge editing framework enhanced with continuous retrieval . they employ an Editing-Aware Learning Strategy and a LawEdit Retriever . |
| Outcome: | a new framework outperforms existing methods for updating legal knowledge in Korean . it maintains robust performance in sequential editing and is qualitatively validated by legal experts. |
KEBAP: Korean Error Explainable Benchmark Dataset for ASR and Post-processing (2023.emnlp-main)
Copied to clipboard
| Challenge: | Conventional evaluation metrics for automatic speech recognition systems produce a singular aggregate score, which is insufficient for understanding specific system vulnerabilities. |
| Approach: | They propose to introduce the Korean Error Explainable Benchmark Dataset for ASR and Post-processing (KEBAP) this method enables a more balanced assessment encompassing speech recognition accuracy and user readability. |
| Outcome: | The proposed method enables a more balanced assessment encompassing speech recognition accuracy and user readability. |
Gap-K%: Measuring Top-1 Prediction Gap for Detecting Pretraining Data (2026.acl-long)
Copied to clipboard
| Challenge: | Existing state-of-the-art methods for pretraining data are largely undisclosed, resulting in ethical and copyright concerns. |
| Approach: | They propose a method that leverages the log probability gap between the top-1 predicted token and the target token, incorporating a sliding window strategy to capture local correlations and mitigate token-level fluctuations. |
| Outcome: | The proposed method outperforms baselines on WikiMIA and MIMIR benchmarks and achieves state-of-the-art performance. |
MMAC: A Multilingual, Multimodal Alignment Framework for Cultural Grounding Evaluation (2026.acl-long)
Copied to clipboard
Weihua Zheng, Zhengyuan Liu, Tanmoy Chakraborty, Weiwen Xu, Xiaoxue Gao, Bryan Chen Zhengyu Tan, Bowei Zou, Chang Liu, Yujia Hu, Xing Xie, Xiaoyuan Yi, Jing Yao, Chaojun Wang, Long Li, Rui Liu, Huiyao Liu, Koji Inoue, Ryuichi Sumida, Tatsuya Kawahara, Fan Xu, Lingyu Ye, Wei Tian, Dongjun Kim, Jimin Jung, Jaehyung Seo, Nadya Yuki Wangsajaya, Pham Minh Duc, Ojasva Saxena, Palash Nandi, Xiyan Tao, Wiwik Karlina, Tuan Luong, Keertana Arun Vasan, Roy Ka-Wei Lee, Nancy F. Chen
| Challenge: | Existing models lack cultural alignment across modalities and languages . a new framework to assess cultural awareness across linguistics and languages is needed . |
| Approach: | They propose a framework that integrates tri-modally aligned cultural benchmarks and a five-dimensional evaluation protocol to assess cross-country awareness disparities. |
| Outcome: | The proposed framework assesses cultural awareness disparities across modalities and languages . it is the first dataset aligned at the input level across text, image, and speech . |
Debiasing Online Preference Learning via Preference Feature Preservation (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent preference learning frameworks for large language models (LLMs) simplify human preferences with binary pairwise comparisons and scalar rewards. |
| Approach: | They propose a preference feature preservation framework that preserves the distribution of human preference features and maps them throughout the online preference learning process. |
| Outcome: | The proposed framework maintains the distribution of human preference features and utilizing such rich signals throughout the online preference learning process. |
No Reader Left Behind: Multi-Agent Summaries Everyone Can Understand (2026.acl-long)
Copied to clipboard
| Challenge: | Existing summarization systems struggle to address diverse linguistic and cognitive barriers among general readers. |
| Approach: | They propose a multi-agent framework that integrates template-based planning with an iterative feedback loop guided by simulated readers and domain expert revision to address comprehension barriers such as unknown terms, missing contexts, and confusing sentences. |
| Outcome: | The proposed framework improves readability and factuality across multiple datasets and human evaluations show that it is more accessible to a wide range of readers. |
Structural Reasoning Improves Molecular Understanding of LLM (2025.acl-long)
Copied to clipboard
| Challenge: | Recent advances in large language models (LLMs) have shown significant performance, approaching human perception levels. |
| Approach: | They propose an approach that sketches molecular structures for reasoning by explicitly incorporating key structural features into the model. |
| Outcome: | The proposed framework improves molecular understanding through extensive experiments. |
EMCEE: Improving Multilingual Capability of LLMs via Bridging Knowledge and Reasoning with Extracted Synthetic Multilingual Context (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods emphasize reformulating queries into English, but fail to incorporate language- and culture-specific grounding that is essential for some queries. |
| Approach: | They propose a framework that extracts query-relevant knowledge from the LLM itself. |
| Outcome: | The proposed framework outperforms existing approaches on four multilingual benchmarks covering diverse languages and tasks. |
Improving Chemical Understanding of LLMs via SMILES Parsing (2025.emnlp-main)
Copied to clipboard
| Challenge: | Molecular string representations such as SMILES and SELFIES are becoming a standard format for applying large language models (LLMs) however, molecular strings follow complex syntactic rules for encoding molecules, which LLMs struggle to interpret. |
| Approach: | They propose a framework that parses SMILES into clean and deterministic tasks to promote graph-level molecular comprehension. |
| Outcome: | The proposed framework improves structural comprehension and competes with the baseline on the Mol-Instructions benchmark. |
Personalized Language Models via Privacy-Preserving Evolutionary Model Merging (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for personalization in language models lack explicit mechanisms for privacy preservation. |
| Approach: | They propose a Privacy-Preserving Model Merging via Evolutionary Algorithms to optimize utility while minimizing privacy risks. |
| Outcome: | The proposed approach outperforms baseline models on the LaMP benchmark and achieves 45% improvement in task performance. |
Embracing Anisotropy: Turning Massive Activations into Interpretable Control Knobs for Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Prior work shows that Large Language Models exhibit highly anisotropic internal representations . prior work shows specialized dimensions capture domain-specific features . |
| Approach: | They propose a simple magnitude-based criterion to identify Domain-Critical Dimensions in a training-free manner. |
| Outcome: | The proposed method outperforms whole-dimension steering in domain adaptation and jailbreaking scenarios. |
Personalized LLM Decoding via Contrasting Personal Preference (2025.emnlp-main)
Copied to clipboard
| Challenge: | Personalization of large language models (LLMs) is becoming increasingly important as they are increasingly deployed in real-world applications. |
| Approach: | They propose a decoding-time approach that leverages the user's implicit reward signal by performing parameter-efficient fine-tuning on user-specific data. |
| Outcome: | The proposed approach improves personalization by an average of 10.57% in ROUGE-L without external reward models or additional training procedures. |
Revisit What You See: Revealing Visual Semantics in Vision Tokens to Guide LVLM Decoding (2026.acl-long)
Copied to clipboard
| Challenge: | Large Vision–Language Models (LVLMs) integrate visual perception with language understanding, but how vision information contributes to the model’s decoding process remains under-explored. |
| Approach: | They propose a simple training-free decoding method that guides text generation in Large Vision–Language Models by Referencing Vision Tokens. |
| Outcome: | The proposed method leverages the semantic information embedded within vision tokens by projecting it into the text token distribution. |
RoAST: Robustifying Language Models via Adversarial Perturbation with Selective Training (2023.findings-emnlp)
Copied to clipboard
Jaehyung Kim, Yuning Mao, Rui Hou, Hanchao Yu, Davis Liang, Pascale Fung, Qifan Wang, Fuli Feng, Lifu Huang, Madian Khabsa
| Challenge: | Several perspectives of robustness for pre-trained language models have been studied independently, but lacking a unified consideration in multiple perspectives. |
| Approach: | They propose a technique to enhance the multi-perspective robustness of LMs by introducing adversarial perturbation while the model parameters are selectively updated upon their relative importance. |
| Outcome: | The proposed technique improves the robustness of LMs by incorporating four perspectives on model robustness. |
infoVerse: A Universal Framework for Dataset Characterization with Multidimensional Meta-information (2023.acl-long)
Copied to clipboard
| Challenge: | Several methods for characterizing datasets based on model-driven meta-information have been developed, but the relationship and complementary effects of these methods have received less attention. |
| Approach: | They propose a framework that captures multidimensional characteristics of datasets by incorporating various model-driven meta-information. |
| Outcome: | The proposed framework outperforms baselines in three real-world applications and can be used in a variety of real-time problems. |
QUAK: A Synthetic Quality Estimation Dataset for Korean-English Neural Machine Translation (2022.coling-1)
Copied to clipboard
| Challenge: | despite its high utility, there are limitations concerning manual QE data creation. |
| Approach: | They propose to generate a Korean-English QE dataset that is fully automatic . they find that the algorithm is more accurate and faster than manual QE . |
| Outcome: | The proposed datasets show that they scale up to 1.58M and 6.58M, respectively, and show that the results are significantly better when compared to the previous datasets. |