Many very smart people have tried and failed. State of the art remains very basic supervised models with hand engineered features. In the markets, data is perma

Many very smart people have tried and failed. State of the art remains very basic supervised models with hand engineered features. In the markets, data is permanently scarce, so these methods don't work well. In the RL problems that DeepMind is solving, data is literally unlimited, and that's the problem space that these methods have been designed for.