Challenges of real-world reinforcement learning

blog.acolyer.org

48 pointsmotivic5 comments

Reinforcement learning without simulation is not feasible for complex problems.

If you can’t simulate, stick to multi-armed bandits.

anyone know a good survey paper about current SOA? i've read sutton&barton, skimmed openai "Spinning Up in Deep RL". anything else out there?

The Deep RL Bootcamp lectures are still good and relevant:

beggars can't be choosers of course but i would prefer something to read rather than something to watch