What I Cannot Control, I Do Not Understand

blog.evjang.com

4 pointsericjang2 comments

Great post A small remark regarding point (2) of the commercial justifications; it's often conflated that RL "requires" Deep Learning (DL), or that RL is Deep RL, which is clearly a false assertion - one often made by groups often seen as "leading" the field - e.g. OpenAI. It should not be necessary for the future of RL to have such a dependence on DL and perhaps the assumption of DL is leading a large section of research into a local optima and somewhat harmful to research and newcomers to the field of RL in general

Appreciate your posts and looking forward to more!

I think that's a fair point, though I think that semantic meaning is often decided by popular vote. If the masses of undergrads excited about AI have decided that RL == "Deep RL" == DQN, then that's the practical definition one should when discussing RL around the water cooler (for instance, "whether RL works in commercial applications").

I am not really aware of any RL (deep or not) being deployed in commerce besides bandit problems used in recommender systems and the strong similarities between shortest-paths / graph planning. Any thoughts?

Reply on news.ycombinator.com