KAIST Introduction to Reinforcement Learning · Project ·
(Solving Korean Four-Ball Carom Billiards with Reinforcement Learning: Meta-Pretrained Initialization under Sparse Rewards)
TBD
Papers I've been working on,
Publication
KAIST Automated Software Testing · Project ·
(ASTraMut: AST-Guided Mutation Testing for Java)
TBD
KAIST Introduction to Artificial Intelligence · Assignment 3 · Pacman Competition Award ·
Capture-the-Flag Pacman with Self-Play Tuned Heuristics
Two-vs-two CTF Pacman team — goal-commit ghost-aware A* offense, alpha-beta minimax defense, and a 42-feature linear evaluator tuned by population-based self-play with held-out zoo anchors. 40/40 on the official grading run; 77.1% over a 792-game external SOTA sweep with zero runtime errors.
KAIST Introduction to Artificial Intelligence · Assignment 2 ·
Multi-Agent Search for Pacman: Reflex, Minimax, and Alpha-Beta
Adversarial multi-agent search on the CS188 Pacman framework — reflex evaluator (food / ghost / scared-ghost terms), n-agent minimax with depth-on-wraparound, and alpha-beta pruning matched to the autograder’s strict-inequality reference, plus a discussion of action-ordering effects on pruning.
KAIST Introduction to Artificial Intelligence · Assignment 1 ·
Graph Search for Pacman: DFS, BFS, UCS, and A*
Classical graph search on the UC Berkeley CS188 Pacman framework — DFS / BFS / UCS / A* with a shared visited-set template. Includes a written reflection on pop-vs-push goal testing, lazy duplicate handling for priority-queue searches, and where each algorithm’s optimality guarantee actually comes from.
UNIST Machine Learning · Final Project Report ·
SKiP: SVM weighted by K-Nearest-Neighbors and class Probability for weakening outliers
Dual-confidence SVM weighting: global probability captures feature outliers; local KNN consistency captures label outliers. Additive (rather than multiplicative) combination yields stable robustness across noise levels (Iris, Wine, Titanic).
Korean Database Conference (KDBC) 2025 ·
Entropy-Guided Adaptive Label Propagation for Location-Aware Graph Clustering
Adaptive label propagation (ALP) for LBSN graph clustering. Real-time entropy-based weighting balances structural vs. spatial information — separates structurally connected but geographically distant cities (e.g. Nashville vs. Atlanta).
UNIST Introduction to Algorithms · Best Paper Award ·
Hylos: Hierarchically Localized Optimization Strategy for TSP
Hybrid TSP solver: k-means clustering + dynamic dispatch between Held-Karp (exact) and Christofides (approximate). ~7× faster than Christofides on mona-lisa100k (under 10s vs. 66s+) without accuracy loss.
UNIST Introduction to Algorithms · Assignment 1 ·
Comparative Study of Twelve Sorting Algorithms
C++ implementation and benchmarking of six classical (Bubble, Selection, Insertion, Quick, Heap, Merge) and six modern (Cocktail Shaker, Comb, Tournament, Library, Intro, Tim) sorts across random / sorted / reverse / partial inputs at 10³–10⁶. Tim and median-of-three Quick scaled cleanly to 10⁶, where Bubble took 3942s vs Tim’s 0.2s.
ICROS (Institute of Control, Robotics, and Systems) 2024 ·
Development of Intuitive Steering Mechanism for Hands-Free Operation of Indoor Shared Mobility
YoungCHA — a chair-shaped indoor shared mobility with hands-free, intuitive steering. Saddle rotation is read by a potentiometer and tracked by a PID-controlled steering motor (STM32F303RE, encoder calibration via a hard-stop button); the throttle is replaced by a kick-to-start scheme. Designed to provide both seated rest and motion in large indoor venues.
Patent
System and Method for Spatially Proximate Community Detection Based on Entropy-Weighted Adaptive Label Propagation
KR 10-2026-0027653 · Filed 2026-02-11