PhD student at OSI LAB of KAIST AI working on large language models.

💌 [email protected] 🌐 itsnamgyu @ { Google Scholar Twitter LinkedIn GitHub }

I’m interested in novel approaches to improve the efficiency and helpfulness of LLMs.

Recently, I investigated the distillation of chain-of-thought (CoT) reasoning from 100B+ LLMs to small language models [C4], and applied this to hate speech detection [C5].

I am actively working on how to improve LLMs, focusing on these two questions: (1) how to handle the variance in computation/modeling capacity requirements of each token, and (2) how to incentivize the model to express/consider its own uncertainty.

• Previous research topics

Education

• Sogang University (Seoul), BS in Computer Science and Engineering Mar 2016 – Aug 2021 • 3.87/4.3 (Summa Cum Laude)
• KAIST (Seoul), MS in Artificial Intelligence Sep 2021 – Aug 2023 • 3.85/4.3 OSI LAB (Advisor: Se-Young Yun)
• KAIST (Seoul), PhD in Artificial Intelligence Sep 2023 – OSI LAB (Advisor: Se-Young Yun, Co-Advisor: James Thorne)

Publications

(Highlighted)

• [C5] Yongjin Yang*, Joonkee Kim*, Yujin Kim*, Namgyu Ho, James Thorne, Se-young Yun. "HARE: Explainable Hate Speech Detection with Step-by-Step Reasoning." Findings of EMNLP 2023. [Paper]
• [C4] Namgyu Ho, Laura Schmid, and Se-Young Yun. "Large Language Models Are Reasoning Teachers." ACL 2023. [Paper] [Twitter] [GitHub]
• [C3] Jaehoon Oh*, Sungnyun Kim*, Namgyu Ho*, Jin-Hwa Kim, Hwanjun Song, and Se-Young Yun. "Understanding Cross-Domain Few-Shot Learning Based on Domain Similarity and Few-Shot Difficulty." NeurIPS 2022. [Paper]
• [C2] Jaehoon Oh*, Sungnyun Kim*, Namgyu Ho*, Jin-Hwa Kim, Hwanjun Song, Se-Young Yun. "ReFine: Re-Randomization before Fine-Tuning for Cross-Domain Few-Shot Learning." CIKM 2022. [Paper]
• [J1] Namgyu Ho*, and Yoon-Chul Kim*. "Evaluation of transfer learning in deep convolutional neural network models for cardiac short axis slice classification." Scientific reports 11 (2021). [Paper]

(Other)

• [J2] Namgyu Ho, and Yoon-Chul Kim. 2022. "Estimation of Cardiac Short Axis Slice Levels with a Cascaded Deep Convolutional and Recurrent Neural Network Model." Tomography 8 (2023). [Paper]
• [C1] Namgyu Ho*, Yoon-Chul Kim*, and Yeon Hyeon Choe. "Cardiac short-axis slice range classification via transfer learning: Evaluation of seven popular deep CNNs.” ISMRM 2019. [Paper]

Experience

• Research Intern @ LG EXAONE Lab (Seoul) (Dec 2023 – Jun 2023)
• Undergraduate Researcher @ Sogang University ECL Lab (Seoul) (Jun 2020 – Aug 2020) Application of reinforcement learning on SSD schedulers. Under the supervision of Prof. Hyuk-Jun Lee.
• SWE Intern / Team Lead @ ZionTech Solutions (SF Bay Area) (Mar 2019 – Jan 2020) Initiated and lead a major effort to migrate the entire frontend of ZionTech’s flagship cloud service “Wavity” to the Angular framework, in production as of summer 2020.
• Undergraduate Researcher @ Samsung Medical Center (Seoul) (2018, 2020 – 2021) Application of transfer learning, deep CNNs, and recursive networks on cardiac MRI. Under the supervision of Prof. Yoon-Chul Kim.

Honors & Activities

• Contributhon 2020 (Aug 2019 – Nov 2020) Korean translation of PyTorch online tutorials
• Naver D2 Campus Fest 2019 2nd place (Jan 2019 – Feb 2019) Open-source project using Python/Django
• Alpha Sigma Nu (2018 – ) The honor society of Jesuit colleges and universities
• Contributhon 2018 (Aug 2018 – Nov 2018) Korean translation of Keras online tutorials w/ KerasKorea
• Microsoft Student Partners (2018/2019) Hosted various tech evangelism events at Sogang University.
• Release (Student Organization) President (Mar 2017 – Aug 2018) CS student organization for software development at Sogang University. Hosted various seminars, hackathons, etc.

Scholarships

• Silicon Valley Data Science Program, Sogang University (Aug 2017) Scholarship for one-month data science program in SF Bay Area
• Academic Excellence Scholarship, Sogang University (Spring 2017)