• If you are citizen of an European Union member nation, you may not use this service unless you are at least 16 years old.

  • You already know Dokkio is an AI-powered assistant to organize & manage your digital files & messages. Very soon, Dokkio will support Outlook as well as One Drive. Check it out today!


Myths of Reinforcement Learning

This version was saved 15 years ago View current version     Page history
Saved by Satinder Singh
on March 21, 2009 at 8:41:01 am

Each myth/misstatement is discussed on its own page

  1.  Large state spaces are hard for RL
  2. RL is slow
  3. RL does not have (m)any success stories since TDgammon
  4. RL does not work well with function approximation
  5. Value function approximation does not work (and so we should do something else - the current

    favorite alternative seems to be policy search)

  6. Non-Markovianness invalidates standard RL methods
  7. POMDPs are hard for RL to deal with
  8. RL is about learning optimal policies


The following old myths are also unfortunately still around and still damaging for the field

  • RL is model-free (or direct)
  • RL is tabula rasa
  • RL is table lookup
  • RL = Q-learning or perhaps TD

Comments (0)

You don't have permission to comment on this page.