About Me
Senior undergraduate at KAIST (School of Computing & Nuclear and Quantum Engineering), interested in computer architecture, systems software, and hardware–software co-design.
Research Interests
- Memory Hierarchy & RTL-Level Design: architectural exploration of memory hierarchy design with interest in RTL-level hardware component development.
- CXL-Based Disaggregated Memory: interest in memory-pressure-driven smart routing in CXL-based disaggregated memory systems and hardware/software co-design.
- Distributed ML State Management: seamless resumption of distributed ML training through consistent multi-process state capture (RNG, optimizer, data loader state).
Preprints
SHIFT: Sigmoid-Based Heuristic Invertible Fitness-Landscape Transformation
2025Jeongjin Han, Seunghoon Sim, Jian Lee, Seongyoon Park — First author
VRAIL: Vectorized Reward-based Attribution for Interpretable Learning
2025Co-author (equal contribution)
Research Experience
Nuclear I&C and Autonomous Operation Lab, KAIST
Winter 2025Undergraduate Researcher — Advisor: Prof. Jonghyun Kim
- Identified system-level bottlenecks under memory and bandwidth constraints across heterogeneous CPU–GPU pipelines.
- Led pipeline redesign to resolve a GPU underutilization bottleneck caused by a CPU-bound simulator; decoupled policy optimization from environment interaction, improving hardware utilization.
- Reverse-engineered the process memory layout of a closed-source Windows VM simulator to extract and inject runtime state variables, enabling system integration without source access or binary modification.
- Designed and implemented a TCP/Docker-based distributed execution infrastructure bridging a Windows VM and a GPU server, managing cross-system state synchronization under memory bandwidth and latency constraints.
Reactor Physics and Transmutation Lab, KAIST
Winter 2024Undergraduate Researcher — Advisor: Prof. Yonghee Kim
- Conducted Monte Carlo neutron transport simulations using OpenMC, analyzing neutron moderation behavior under Thermal Scattering Law (TSL) models from the ENDF/B-VIII.1 nuclear data library.
- Investigated computational and memory characteristics of large-scale Monte Carlo simulation workloads, and analyzed scalability behavior under parallel execution.
Tech Stack
Languages
C++RustPythonMATLABScalaVerilog
Tools & Frameworks
PyTorchgem5DockerHuggingFaceLaTeXGitLinux