The idea of MAZE is to train machine learning models in different environments and let them move around based on conditions. It's a bit like walking inside a maze. But this maze is directional, we design different environments to guide the evolving different types of machine learning models. These environments are connected together, if one model is performing well, it may get promoted into the next environment. So if you see each environment as a maze, the whole is like the maze of maze, or layers of maze.