
Yonsei Biochem & CS
Sunghyun Park
Biography.
Hello, my name is Sunghyun Park, but I go by Henry in English. I am an undergraduate student majoring in Biochemistry and Computer Science at Yonsei University in Seoul, South Korea.
My research interests lie at the intersection of machine learning for science, optimization and the modeling of complex natural systems. I focus on world model that blends data-driven learning with established domain models to build accurate, robust models. I enjoy developing hybrid approaches that incorporate physical priors, perform structure-aware regularization, and leverage system identification or data assimilation to close the loop between observation, simulation, and prediction.
In parallel, I investigate scientific reasoning in large language models as practical tools for research—designing workflows that pair LLMs with verifiers, external tools, and alignment techniques to improve faithfulness, calibration, and safe decision support. On the application side, I explore representation learning for complex biological data—especially cellular and microscopy images—using self-supervised objectives, segmentation and tracking pipelines, and geometry-aware embeddings to extract interpretable structure from noisy measurements. Across these directions, my goal is to connect theoretical insight with end-to-end practice: principled algorithms that are transparent and reliable, yet deliver real utility in AI-augmented scientific discovery and computational biology.
Education.
-
Mar. 2020 - Present
Yonsei University, Seoul, South Korea
Undergraduate Student in Biochemistry & Computer Science (Double Major)
- GPA: 3.91/4.3
- Oct. 2022 - Jul. 2024 : Compulsory Military Service, Served as a sergent in R.O.K Air Force.
Projects.

My Academic Homepage
This Project is the website you are currently watching. The site includes features such as Gallery and Blog, in particular, the Blog supports search and parses Markdown files, allowing category-based search in a tree structure.
Member: Sunghyun Park

LLM Course Project : Prompt Optimization
- Implemented a prompt optimization loop combining OPRO with an evolutionary algorithm (selection, crossover, mutation) - Used an LLM scorer to evaluate candidates and drive self‑rewriting and selection - Evaluated on GSM8K with Llama-3.1-8B-Instruct and Mistral-7B-Instruct.
Member: Sunghyun Park

Force‑Driven Mechano‑Typing of B Cells - Multi-channel image analysis of CD40-CD40L interaction using autoencoder
We combined multi‑threshold molecular tension probes and three‑channel microscopy (DIC, RICM, TIRF) with a deep autoencoder to encode single‑cell images into 32‑D biophysical representations. This framework reveals how mechanical forces and CD40L mutations jointly shape B‑cell morphology and adhesion.
Members: Sunghyun Park, Yelim Jeon, Hyunkyu Choi

Reducing WebOS Boot Latency
Using ftrace and kernel instrumentation, we captured boot‑time function traces to identify hotspots and page faults. By reorganizing code sections via a custom linker script and optimizing the ELF binary layout, we reduced kernel page faults and improved boot times for embedded devices.
Members: Sunghyun Park, Jonghyun Hwang, Sungyoon Lim, Changwon Kim
Topics of Interest.
Optimization Theory and Training Dynamics in Deep Learning
I am deeply interested in theoretical analysis of optimization algorithms such as Gradient Descent and SGD, and their dynamics in training modern architectures including transformers and diffusion models. I also explore optimal transport and Wasserstein gradient flows as a way to better understand learning trajectories.
Inference Strategies and Structural Understanding of LLMs
My focus lies in decoding strategies like speculative decoding, self-refinement, and verifier-based inference (e.g., Monte Carlo tree search). Rather than mere usage, I aim to understand how large language models function and what emergent behaviors arise from their training dynamics.
Representation Learning for Biological Image Analysis
I apply deep learning to analyze cellular and biological image data, exploring unsupervised feature learning through AutoEncoders and VAEs. My work includes clustering multi-channel microscopy data and understanding latent structures relevant to biological conditions or experimental forces.
Interpretable Deep Learning and Semantic Understanding of Models
I am fascinated by understanding how deep networks make decisions. I explore attention maps, occlusion-based interpretation, and latent space visualization to uncover the semantic structure learned by models and connect them to domain-relevant insights.
Experience.
My main research topic in Mechonbiology Lab was the investigation of Deep-Learning approach to make connection with AI and Biology, which involves multi-cell image analysis using compuatational methods.
In PoolC, I mainly studied and implemented visual data generation models based on 3D morphable face models and neural renderers, specifically aiming to achieve better-performing expression/identity swapping between different images or frames.
Awards & Honors.

Honors Award, Yonsei - Spring 2025
Awarded to the top 10% students in Yonsei Students

Honors Award, Yonsei - Spring 2021
Awarded to the top 10% students in Yonsei Students
Contact.
- psh040@yonsei.ac.kr
- +82-10-9115-0823