SOUL - Currency War Simulation

Multi-agent reinforcement learning project simulating currency wars to derive new economical and political strategies

Project Overview

SOUL is an ambitious research project funded by the Korea Foundation for Advanced Studies (KFAS) that uses multi-agent reinforcement learning to simulate currency wars and derive novel economical and political strategies.

Goal

The primary objective is to simulate complex economic scenarios involving currency competition between multiple agents, allowing us to:

  • Understand emergent behaviors in currency markets
  • Derive new strategic insights for economic policy
  • Demonstrate the power of cooperative AI through communication

Technical Approach

Multi-Agent Reinforcement Learning

  • Implemented game-playing AI agents using Deep Q-Learning
  • Agents learn to cooperate and compete through communication channels
  • Utilized Long Short-Term Memory (LSTM) networks for sequential decision-making

Infrastructure

  • Deployed on Google Cloud Platform for scalable training
  • Managed distributed training across multiple compute instances
  • Optimized for large-scale simulations with thousands of interactions

Key Achievements

  • Funding: Secured 10 million KRW (~$10,000 CAD) in project funding
  • Resource Management: Successfully allocated resources for computing infrastructure, expert mentoring, accommodation, and educational materials
  • Knowledge Transfer: Communicated complex AI concepts including LSTM to team members with non-technical backgrounds
  • Ongoing Research: Project continues to evolve with new findings since February 2022

Technical Stack

  • Languages: Python
  • Frameworks: PyTorch, PettingZoo (for multi-agent environments)
  • Libraries: NumPy, Pandas, Matplotlib
  • Infrastructure: Google Cloud Platform

Team & Role

As Project Head, I:

  • Lead the technical architecture and implementation
  • Manage team coordination and resource allocation
  • Communicate with stakeholders and mentor team members
  • Drive research direction and publication efforts
This project demonstrates the intersection of economics, game theory, and cutting-edge AI techniques.

Project Status: Finished (Feb 2022 - Jan 2023)
Organization: Korea Foundation for Advanced Studies (KFAS)
Location: Seoul, South Korea