Inspiration
Machine learning requires lots of compute power. It can take days to train a model.
JH has lots of computers in it, many of them sitting doing nothing.
Can we leverage these unused machines to help train our models?
What it does
We create a pool of machines to run a bracket-style tournament. Each machine trains the model it is given, then the models compete against each other. The losing models are dropped and the process repeats. This goes on until only the best model remains.
The model used is a CNN which is learning to play Connect 4
Built With
- python
- tensorflow
- typescript
Log in or sign up for Devpost to join the conversation.