Holds the functions asociated with actual NN such as forward pass and mutateĭifferent NN models (1 and 2 are Stage 1, 3 is Stage 3) Holds the main training loop and functions asociated with mutation of networks, elimination and cloning of networks Read more about Google DeepMind's AlphaGo Zero project here Module descriptionįollowing scripts are of most importance: The documentary also contributed to the motivation behind the project. The motivation behind the project is the recent advancement of AI using Reinforcement Learning, and specially the AlphaGo and AlphaGo Zero created by Google DeepMind.

You may read the other suggested final projects for the course here Motivation Please have a look at the PDF files TDT4265_Final_project_proposal.pdf and Learning to play Go - Final presentation.pdf to see our final results

The rules used in the implementation is Chinese rules, also known as area rules Final project in TDT4265 - Computer Vison and Deep Learning at NTNU using Deep Reinforcement Learning with Genetic Algorithm to train an AI to play the game of Go on a 9x9 board.

Betago algorithm