Reinforcement Learning with a Neural Network and multiple players

[Zomis]

One of my big dreams is to make good AIs that can learn to play almost any game. Therefore, take some steps towards that.

The goal of this issue is to implement Reinforcement Learning AI using Neural Network for any games.

Good candidates for Reinforcement Learning:

Super simple one-player games:

• Classic Maze-game with a goal and one or more holes that give penalty.
• Classic 1234 slots game where you should do each action one time to get reward. Doing an action twice gives penalty.

"Simple" two player games:

• Nim
• TTTUpgrade, simple two player games
• Quixo, TTT3D, UTTT, Reversi, Connect4, TTT, UR, Pentago

Trickier two player games:

• Backgammon
• Battleship?

"Simple" multiplayer games:

• Artax, available in multiplayer for 2-4 players

Mind games:

• Skull (check past X rounds to figure out future, kinda like how the good Rock-Paper-Scissors AIs work)