Projects
Reversi AI Agent (GitHub Community's Reversi)
A reinforcement learning agent trained to play Reversi, capable of defeating human opponents after sufficient self-play and training.
- Implemented using Python and PyTorch
- Trained locally with a Deep Q-Network (DQN) architecture via self-play
- Used GitHub Actions to automate README updates via issue-triggered workflows
gamealloc
A Python package for modeling and solving resource allocation problems using game-theoretic frameworks. It implements classical algorithms such as Sequential Priority and Top Trading Cycle to handle matching and trading markets in multi-agent scenarios.
- Designed APIs for checking Pareto efficiency and strategy-proofness across allocation instances
- Implemented preference manipulation tools and benchmarking routines
- Packaged for easy reuse, installable via pip and documented with example notebooks
100 Prisoners Problem Simulator
A Python tool simulating the classic "100 prisoners and 100 boxes" puzzle under two strategies: random picks or cycle-following strategy. Designed to visualize success rates and demonstrate the counterintuitive probability behind the optimal solution.
- Implemented both random-choice and cycle-following algorithms in Python
- Simulates repeated runs to estimate survival probability under each strategy
- Developed a Streamlit app to simplify parameter setting and result visualization
Cost-sensitive-Feature-Selection-for-Support-Vector-Machines Public
A Python implementation of a cost-sensitive feature selection method for Support Vector Machines (SVM), balancing classification accuracy with feature acquisition cost. The approach minimizes total feature cost while satisfying true-positive and true-negative rate constraints.
- Phase-I: Uses integer linear programming (ILP) with L1-norm SVM to select low-cost feature subsets
- Phase-II: Trains final SVM with linear or radial kernel under cost-aware constraints
- Evaluated using k-fold cross-validation to guarantee performance across datasets