SooHwan Eom

Daejeon, Korea

I am a Ph.D. student in Electrical Engineering at KAIST, advised by Prof. Chang D. Yoo. I received my B.S. (2022) and M.S. (2024), both in Electrical Engineering from KAIST.

My research focuses on large language model (LLM) alignment and reasoning, multimodal learning, and self-supervised speech representation learning. Recent projects include parameter-efficient adaptation of foundation models and multimodal language models. More broadly, I study principled methods for adapting and improving foundation models under limited, noisy, or structured supervision — with the long-term goal of building self-improving, agentic AI systems that reason reliably in complex domains.

news

Jun 2026	Token-level Response-visual Attention Guidance for Multimodal LLMs Knowledge Distillation got accepted to ECCV 2026.
Jun 2026	Transcript-Free Flow-Matching Text-to-Speech via Speech Feature Conditioning got accepted to Interspeech 2026.
May 2026	Decomposed On-Policy Distillation for Vision-Language Reasoning: Steering Gradients for Visual Grounding got accepted to ICML 2026 (Spotlight, top 2.2%).
Feb 2026	PDCR: Perception-Decomposed Confidence Reward for Vision-Language Reasoning got accepted to CVPR 2026.
Sep 2025	Our review on wearable blood pressure sensors and machine learning algorithms for blood pressure estimation was published in Nature Reviews Cardiology.

selected publications

INTERSPEECH

Transcript-Free Flow-Matching Text-to-Speech via Speech Feature Conditioning

SooHwan Eom, Hee Suk Yoon, Eunseop Yoon, Mark Hasegawa-Johnson, and Chang D. Yoo

In Proceedings of INTERSPEECH, 2026

arXiv Bib

@inproceedings{eom2025transcript,
  title = {Transcript-Free Flow-Matching Text-to-Speech via Speech Feature Conditioning},
  author = {Eom, SooHwan and Yoon, Hee Suk and Yoon, Eunseop and Hasegawa-Johnson, Mark and Yoo, Chang D.},
  booktitle = {Proceedings of INTERSPEECH},
  year = {2026},
  archiveprefix = {arXiv},
  primaryclass = {eess.AS}
}

ICML

Decomposed On-Policy Distillation for Vision-Language Reasoning: Steering Gradients for Visual Grounding

Hee Suk Yoon, Eunseop Yoon, Jaehyun Jang, SooHwan Eom, Ji Woo Hong, Mark Hasegawa-Johnson, Qi Dai, Chong Luo, and Chang D. Yoo

In Proceedings of the International Conference on Machine Learning (ICML), 2026

Spotlight Paper arXiv Bib

Selected as a Spotlight (top 2.2%) at ICML 2026

@inproceedings{yoon2026decomposed,
  title = {Decomposed On-Policy Distillation for Vision-Language Reasoning: Steering Gradients for Visual Grounding},
  author = {Yoon, Hee Suk and Yoon, Eunseop and Jang, Jaehyun and Eom, SooHwan and Hong, Ji Woo and Hasegawa-Johnson, Mark and Dai, Qi and Luo, Chong and Yoo, Chang D.},
  booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
  year = {2026},
  doi = {10.48550/arXiv.2606.00564},
  url = {https://arxiv.org/abs/2606.00564},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV}
}

INTERSPEECH

SiamCTC: Learning Speech Representations through Monotonic Temporal Alignment

SooHwan Eom, Mark Hasegawa-Johnson, and Chang D. Yoo

In Proceedings of INTERSPEECH, 2025

Paper arXiv Bib

@inproceedings{eom2025siamctc,
  title = {{SiamCTC}: Learning Speech Representations through Monotonic Temporal Alignment},
  author = {Eom, SooHwan and Hasegawa-Johnson, Mark and Yoo, Chang D.},
  booktitle = {Proceedings of INTERSPEECH},
  year = {2025},
  doi = {10.21437/Interspeech.2025-2746},
  url = {https://doi.org/10.21437/Interspeech.2025-2746},
  archiveprefix = {arXiv},
  primaryclass = {eess.AS}
}

EMNLP

Query-based Cross-Modal Projector Bolstering Mamba Multimodal LLM

SooHwan Eom, Jay Shim, Gwanhyeong Koo, Haebin Na, Mark A. Hasegawa-Johnson, Sungwoong Kim, and Chang D. Yoo

In Findings of the Association for Computational Linguistics: EMNLP 2024, Nov 2024

Paper Bib

@inproceedings{eom2024query,
  title = {Query-based Cross-Modal Projector Bolstering Mamba Multimodal {LLM}},
  author = {Eom, SooHwan and Shim, Jay and Koo, Gwanhyeong and Na, Haebin and Hasegawa-Johnson, Mark A. and Kim, Sungwoong and Yoo, Chang D.},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2024},
  month = nov,
  year = {2024},
  address = {Miami, Florida, USA},
  publisher = {Association for Computational Linguistics},
  pages = {14158--14167},
  doi = {10.18653/v1/2024.findings-emnlp.827},
  url = {https://aclanthology.org/2024.findings-emnlp.827/}
}

ACL

TLCR: Token-Level Continuous Reward for Fine-grained Reinforcement Learning from Human Feedback

Eunseop Yoon^*, Hee Suk Yoon^*, SooHwan Eom^*, Gunsoo Han, Daniel Nam, Daejin Jo, Kyoung-Woon On, Mark Hasegawa-Johnson, Sungwoong Kim, and Chang D. Yoo

In Findings of the Association for Computational Linguistics: ACL 2024, 2024

Paper Bib

@inproceedings{yoon2024tlcr,
  title = {{TLCR}: Token-Level Continuous Reward for Fine-grained Reinforcement Learning from Human Feedback},
  author = {Yoon, Eunseop and Yoon, Hee Suk and Eom, SooHwan and Han, Gunsoo and Nam, Daniel and Jo, Daejin and On, Kyoung-Woon and Hasegawa-Johnson, Mark and Kim, Sungwoong and Yoo, Chang D.},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2024},
  series = {Findings of ACL},
  pages = {14969--14981},
  publisher = {Association for Computational Linguistics},
  year = {2024},
  doi = {10.18653/v1/2024.findings-acl.889},
  url = {https://aclanthology.org/2024.findings-acl.889/}
}

ICASSP

AdaMER-CTC: Connectionist Temporal Classification with Adaptive Maximum Entropy Regularization for Automatic Speech Recognition

SooHwan Eom, Eunseop Yoon, Hee Suk Yoon, Chanwoo Kim, Mark Hasegawa-Johnson, and Chang D. Yoo

In IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024

Paper Bib

@inproceedings{eom2024adamer,
  title = {{AdaMER-CTC}: Connectionist Temporal Classification with Adaptive Maximum Entropy Regularization for Automatic Speech Recognition},
  author = {Eom, SooHwan and Yoon, Eunseop and Yoon, Hee Suk and Kim, Chanwoo and Hasegawa-Johnson, Mark and Yoo, Chang D.},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages = {12707--12711},
  publisher = {IEEE},
  year = {2024},
  doi = {10.1109/ICASSP48485.2024.10446721},
  url = {https://doi.org/10.1109/ICASSP48485.2024.10446721}
}