Reinforcement learning is a very generic and widely applicable framework. If we had any way to solve RL problems, we'd use it everywhere. The way convnets are used everywhere in CV nowadays. The fact that RL is nowhere to be seen in the industry is pretty telling...
-
Maybe. Just because a problem can be seen as an RL problem, this doesn't necessarily mean that this is the most productive way of seeing it. Many problems that could be RL might be better seen as something else (planning, optimization, solving supervised learning, etc).
Isn't that primarily because we know how to solve these other categories of problems (while we don't know how to do RL)? Everything else being equal, RL seems like the simplest, most intuitive way to cast most learning problems.
-
-
RL throws out a lot of information that you have free access to though, and a lot of the things that improve on it or work better do so simply by making better use of that information (intermediate state transitions, ability to achieve alternate goals, etc).
-
Doesn't that mainly come from your modelization and how specific application use RL rather than a weakness in the framework? State is supposed to be sufficient information: no need to know the past, just the present.
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.