Reinforcement Learning

We developed a model that enables an AI to play “4 in a row” using only reinforcement learning. This means that the AI’s move is determined solely by the output of the model, without the use of minimax, tree search, or any other traditional algorithms (unlike software such as AlphaGo). Our goal was to investigate the capabilities of neural networks, rather than creating the strongest AI possible.

You can evaluate our success for yourself by playing our demo (realised with TFLite), here or in the snippet below.

👉 Read our introduction on reinforcement learning

🐱 Github: github.com/itsadeepizza/4inarow