MuZero is a new evolution of the AlphaZero machine learning algorithm — only this time it can play games without prior knowledge of the rules.
No rules is important because of, y’know, the real world. From Nature (link below):
“in real-world problems, the dynamics governing the environment are often complex and unknown.”
In other words, in many real world applications the exact rules of what’s going on simply aren’t clear. They have to be intuited.
This is an explainer of how MuZero builds intuition. It’s fairly accessible as explanations of advanced mathematics go, but I won’t pretend to have really grokked it.
What fascinates me is the way that the mental model here is that of intuition, which kind of nails it. I’m a sucker for research well explained to (relative) laypeople. Well done, researchers!
Links: