Continual Learning in Reinforcement Learning

How to Play

To play, simply run the game.py file. There are no additional installations or dependencies required, you just need python, so it's easy to get started. The entire project is designed to work out of the box, so you can start playing right away without worrying about setting up external libraries or tools.

python game.py

Project Objectives

The main objective of this project is to investigate the concept of continual learning within the field of reinforcement learning (RL) applied to board games, specifically to the game Squadro and its variants. The key objectives include:

• Develop a functional version of the board game Squadro that serves as an experimental platform.
• Apply continual learning techniques to enable the agent to learn continuously over multiple games, without forgetting previously acquired knowledge.
• Evaluate how the use of continual learning improves the agent's ability to adapt to variations of the game, such as rule changes or board size adjustments.
• Conduct experiments to measure the agent's performance and the computational complexity of the system under different continual learning conditions.
• Explore how continual learning can optimize game strategies, allowing the agent to improve over time without requiring complete retraining.

---

Board Game Development 🎲

Pending Tasks

• Develop a first functional version of the board game.
• Add color to the game pieces.
• Create a tutorial for players.
  • Basic tutorial.
  • Examples of moves for the tutorial.

---

Reinforcement Learning 🤖

Pending Tasks

• Apply a RL algorithm to the game.
• Identify the most relevant papers in the RL literature.

---

Experiments and Analysis 🧪

Pre-training Models

• Question: Can pre-trained models play the game Squadro?
• Hypothesis: It should be investigated whether pre-trained models are capable of learning and executing efficient moves in Squadro.
• Experiment: (To be defined)
• Conclusion: (To be defined)

---

Rule Modification

• Question: If we modify the board size, how does it affect the computational complexity of the game?
• Hypothesis: Increasing or reducing the board size may alter the algorithmic complexity of the game, possibly simplifying it.
• Experiment: (To be defined)
• Conclusion: (To be defined)

---

Other Ideas

• Is it feasible to use the Minimax algorithm in this context?
• Train a model using historical game data to improve its performance.

---

This project is under active development, aiming to integrate artificial intelligence algorithms that allow the game strategies to learn and improve automatically. Additionally, experiments will be conducted to deepen the understanding of the system’s computational complexity.