Hi everyone,
I trained an AI to play Mandala using the same self-play reinforcement learning approach behind AlphaZero. No human game data, no hand-coded rules. The bot learns entirely by playing against itself.
It's still early in training and actively improving. You can play against it right now, free, no signup:
https://mandala-rl-production.up.railway.app
I'm looking for feedback from experienced Mandala players. Does it play intelligently? What obvious mistakes does it make? Is the difficulty level interesting?
If you're curious about the technical side: it uses a ResNet policy/value network with MCTS, trained through self-play on GPU. Happy to answer questions.
Thanks for trying it out.
I trained an AI to play Mandala using the same self-play reinforcement learning approach behind AlphaZero. No human game data, no hand-coded rules. The bot learns entirely by playing against itself.
It's still early in training and actively improving. You can play against it right now, free, no signup:
https://mandala-rl-production.up.railway.app
I'm looking for feedback from experienced Mandala players. Does it play intelligently? What obvious mistakes does it make? Is the difficulty level interesting?
If you're curious about the technical side: it uses a ResNet policy/value network with MCTS, trained through self-play on GPU. Happy to answer questions.
Thanks for trying it out.