Suppose we want to maximize reward, but we only get a couple bits of reward data every few hundreds/thousands of actions, whereas we get…
In modern ML, representation learning is the art of trying to find useful abstractions, embodied as encoding networks. We can learn…