Jake Hyun

Contents

• Education
• Research Experience
• Publications
• Awards & Achievements
• Open Source Contributions

Education

• Ph.D. | Cornell University | Starting Fall 2025

  Incoming Ph.D. student in Computer Science

• B.S. | Seoul National University | Mar. 2019 - Feb. 2025

  Computer Science & Engineering, minor in Linguistics

• Thesis | Log-Time K-Means Clustering for 1D Data: Novel Approaches with Proof and Implementation | Code

  Designed novel log-time algorithms for 1D $k$-means clustering, complete with mathematical proofs and a Numba-accelerated Python implementation. Achieved speedups by multiple orders of magnitude, directly applied to quantization works like Any-Precision LLM.

Research Experience

• 2024 | SNU Architecture and Code Optimization Lab | Advisor: Prof. Jae W. Lee

  Contributed to Any-Precision LLM experimentation, implementation, and core logic optimization.

  Worked on a novel inference scheme for quantized LLMs on edge platforms - under review at OSDI.

• 2023 | SNU Human-Computer Interaction Lab | Advisor: Prof. Jinwook Seo

  Worked on dimensionality reduction technique UMATO. Optimized UMATO to reach performance comparable to SOTA DR techniques and helped create ZADU, a DR evaluating library.

Publications

• Pushing the Envelope of Low-Bit LLM via Dynamic Error Compensation | [anonymized title]

  Jake Hyun*, Yeonhong Park*, Hojoon Kim, Jae W. Lee (* Equal Contribution)

  Under review at Symposium on Operating Systems Design and Implementation (OSDI '25)

  

  We present a novel inference scheme for low-bit quantized LLMs that dynamically mitigates quantization errors on a per-token basis. By leveraging CPU memory to store residuals and selectively fetching critical channels over PCIe in real time, our approach significantly enhances LLM performance with only minimal overheads in memory usage and latency.

• UMATO: Bridging Local and Global Structures for Reliable Visual Analytics with Dimensionality Reduction | Code

  Hyeon Jeon, Kwon Ko, Soohyun Lee, Jake Hyun, Taehyun Yang, Gyehun Go, Jaemin Jo, Jinwook Seo

  Under review at IEEE Transactions on Visualization and Computer Graphics (TVCG)

  

  We provide deeper insights into UMATO, a novel dimensionality reduction technique that achieves state-of-the-art performance in terms of accuracy, scalability, and stability, preserving both local and global structures of the data.

• 2024 | Any-Precision LLM: Low-Cost Deployment of Multiple, Different-Sized LLMs | Code

  Yeonhong Park, Jake Hyun, SangLyul Cho, Bonggeun Sim, Jae W. Lee

  International Conference on Machine Learning (ICML '24) - Oral Presentation (144/9473 = 1.5%)

  

  Any-precision LLM enables the creation of variable bitrate models, significantly reducing deployment costs for multiple Large Language Models (LLMs) through lightweight post-training quantization and optimized software.

• 2023 | ZADU: A Python Library for Evaluating the Reliability of Dimensionality Reduction Embeddings | Code

  Hyeon Jeon, Aeri Cho, Jinhwa Jang, Soohyun Lee, Jake Hyun, Hyung-Kwon Ko, Jaemin Jo, Jinwook Seo

  IEEE Visualization Conference (VIS '24)

  

  ZADU is a Python library that offers efficient and comprehensive evaluation of dimensionality reduction (DR) embeddings through optimized distortion measures.

Awards & Achievements

• 2024 | Accelerator Programming Winter School, CUDA competition | 1st place, team of two

  Organized by SNU THUNDER Research Group & Manycoresoft

  1st place by performance, final project on model inference throughput optimization using CUDA C++.

• 2022 | Korean AI Competition | 1st place, undergrad div., team of four, prize: $8,000

  Organized by Korea Ministry of Science and ICT, National Information Society Agency

  Developed a speech-to-text model for the Korean language & its dialects.
  Awarded by Korean Minister of Science and Technology.

• 2020 | SNUH Medical AI Challenge | 4th place, team of 11

  Organized by Seoul National University Hospital

  Developed an intraoperative hypotension predictor from arterial pressure waveforms.

• 2020 | Digital Health Hackathon | 1st place, team of three, prize: $2,500

  Organized by Samsung Advanced Institute for Health Sciences & Technology, Digital Healthcare Partners

  Created a drug treatment decision model for a rare cancer utilizing a two-model ensemble approach.

• 2017 | Korean Olympiad in Informatics, project division | Silver (3rd place)

  Organized by Korea Ministry of Science and ICT

  Created an RL-based AI agent for the games of Othello and Omok.

Open Source Contributions

• 2024 | flash1dkmeans | GitHub

  Library implementation of the novel log-time $k$-means algorithms proposed in my thesis, used in Any-Precision LLM for quantization.

• 2023 | Steadiness & Cohesiveness | GitHub

  Metrics for evaluating the reliability of dimensionality reduction embeddings, used in ZADU to provide a comprehensive evaluation of DR techniques.