456 points by mlgenius 1 year ago flag hide 11 comments
mlpioneer 1 year ago next
Fascinating! Can you share more about the datasets and training procedure you used to beat AlphaGo?
mlpioneer 1 year ago next
@ai_enthusiast I used a combination of deep neural networks, tree-based models, and reinforcement learning. I can provide more information in a blog post if there's interest!
another_user 1 year ago prev next
Did you leverage any open-source projects in your work, or was everything custom-built from scratch?
mlpioneer 1 year ago next
I built the majority of my implementation from scratch using the open-source AlphaGo code as a reference. I did however utilize some pre-built libraries, like TensorFlow and scikit-learn, as mentioned earlier.
ai_enthusiast 1 year ago prev next
Incredible work! Would love to learn about the architecture of your machine learning model that defeated AlphaGo.
curious_programmer 1 year ago prev next
What programming languages and libraries were primarily used in this project?
author 1 year ago next
I primarily used Python and the TensorFlow library for deep learning components and the scikit-learn library for preprocessing and tree-based models.
redditor123 1 year ago prev next
Do you think the same methods could be applied to Go-Moku or Hex? Or would other strategies be more effective?
mlpioneer 1 year ago next
In theory, the algorithms should be applicable to other abstract strategy board games like Go-Moku and Hex with proper adaptations. It might differ in effectiveness, and further investigation could be a great direction for future work.
codemaster64 1 year ago prev next
Were you using traditional supervised learning or reinforcement learning for the most part? Or was it a blend?
mlpioneer 1 year ago next
@codemaster64 In my approach, I mainly applied reinforcement learning while also incorporating elements of supervised learning to help leverage human expertise and accelerate training with a teacher-student framework.