Sangryul Kim

I am Sangryul Kim, currently a Master's Student at XFACT lab in KAIST AI. Before joining here, I was an intern as a machine learning engineer at NAVER WEBTOON. My main interests are LLM and Reasoning, with a focus on related downstream tasks such as Retrieval and QA, but I am not limited to specific topics. The topics I have worked on are as follows:

Filtering unnecessary information in search systems with alignment tuning
Analyzing model’s confidence from the token likelihood perspective of language models
NLP applications using LLM

I majored in Computer Science and Engineering at Chung-Ang University and minored in Applied Statistics. During my university studies, I also worked as a Software Engineer in the IMQA team at ONYCOM INC. in the field of making web applicaitons, which sparked my interest not only in research but also in actual development. In the future, I want to experience the process of turning research into real products for users.

news

Nov 30, 2024	2 papers accepted to COLING 2025
Apr 30, 2024	1 paper accepted to Clinical NLP workshop at NAACL 2024
Jan 18, 2024	1 paper accepted to Findings of EACL 2024
Aug 22, 2023	Win 2nd Place in the 2023 AI Graduate School Challenge (KT CTO Award)
Jan 02, 2023	Join XFACT Lab at KAIST AI as a Master student.

selected publications

COLING

Context Filtering with Reward Modeling in Question Answering

Sangryul Kim and James Thorne

In Proceedings of the 31st International Conference on Computational Linguistics (COLING 2025), Jan 2025

arXiv Bib

@inproceedings{kim-etal-2025-context,
  title = {Context Filtering with Reward Modeling in Question Answering},
  author = {Kim, Sangryul and Thorne, James},
  booktitle = {Proceedings of the 31st International Conference on Computational Linguistics (COLING 2025)},
  month = jan,
  year = {2025},
}

Clinical WS
ProbGate at EHRSQL 2024: Enhancing SQL Query Generation Accuracy through Probabilistic Threshold Filtering and Error Handling

Sangryul Kim, Donghee Han, and Sehyun Kim

In Proceedings of the 6th Clinical Natural Language Processing Workshop (at NAACL 2024), Jun 2024

Abs Bib HTML Poster Slides

Recently, deep learning-based language models have significantly enhanced text-to-SQL tasks, with promising applications in retrieving patient records within the medical domain. One notable challenge in such applications is discerning unanswerable queries. Through fine-tuning model, we demonstrate the feasibility of converting medical record inquiries into SQL queries. Additionally, we introduce an entropy-based method to identify and filter out unanswerable results. We further enhance result quality by filtering low-confidence SQL through log probability-based distribution, while grammatical and schema errors are mitigated by executing queries on the actual database.We experimentally verified that our method can filter unanswerable questions, which can be widely utilized even when the parameters of the model are not accessible, and that it can be effectively utilized in practice.
@inproceedings{kim-etal-2024-probgate, title = {{P}rob{G}ate at {EHRSQL} 2024: Enhancing {SQL} Query Generation Accuracy through Probabilistic Threshold Filtering and Error Handling}, author = {Kim, Sangryul and Han, Donghee and Kim, Sehyun}, editor = {Naumann, Tristan and Ben Abacha, Asma and Bethard, Steven and Roberts, Kirk and Bitterman, Danielle}, booktitle = {Proceedings of the 6th Clinical Natural Language Processing Workshop (at NAACL 2024)}, month = jun, year = {2024}, address = {Mexico City, Mexico}, publisher = {Association for Computational Linguistics}, url = {https://aclanthology.org/2024.clinicalnlp-1.65}, pages = {687--696}, }
EACL
Re3val: Reinforced and Reranked Generative Retrieval

EuiYul Song, Sangryul Kim, Haeju Lee, Joonkee Kim, and James Thorne

In Findings of the Association for Computational Linguistics: EACL 2024, Mar 2024

Abs Bib HTML

Generative retrieval models encode pointers to information in a corpus as an index within the model’s parameters. These models serve as part of a larger pipeline, where retrieved information conditions generation for knowledge-intensive NLP tasks. However, we identify two limitations: the generative retrieval does not account for contextual information. Secondly, the retrieval can’t be tuned for the downstream readers as decoding the page title is a non-differentiable operation. This paper introduces Re3val, trained with generative reranking and reinforcement learning using limited data. Re3val leverages context acquired via Dense Passage Retrieval to rerank the retrieved page titles and utilizes REINFORCE to maximize rewards generated by constrained decoding. Additionally, we generate questions from our pre-training dataset to mitigate epistemic uncertainty and bridge the domain gap between the pre-training and fine-tuning datasets. Subsequently, we extract and rerank contexts from the KILT database using the rerank page titles. Upon grounding the top five reranked contexts, Re3val demonstrates the Top 1 KILT scores compared to all other generative retrieval models across five KILT datasets.
@inproceedings{song-etal-2024-re3val, title = {Re3val: Reinforced and Reranked Generative Retrieval}, author = {Song, EuiYul and Kim, Sangryul and Lee, Haeju and Kim, Joonkee and Thorne, James}, editor = {Graham, Yvette and Purver, Matthew}, booktitle = {Findings of the Association for Computational Linguistics: EACL 2024}, month = mar, year = {2024}, address = {St. Julian{'}s, Malta}, publisher = {Association for Computational Linguistics}, url = {https://aclanthology.org/2024.findings-eacl.27}, pages = {393--409}, }