Abstract:
My paper in the main portion of the conference deals with fitted value iteration or Q-learning for offline problems, {em i.e.}, those where we have a model of the environment so that we can examine arbitrary transitions in arbitrary order. The same techniques also allow us to do Q-learning for an online problem, {em i.e.}, one where we have no model but must instead perform experiments inside the MDP to gather data. I will describe how.
copied = false, 2000);
">
@workshop{Gordon-1995-16193,
author = {Geoffrey Gordon},
title = {Online Fitted Reinforcement Learning},
booktitle = {Proceedings of ICML '95 Value Function Approximation in Reinforcement Learning Workshop},
year = {1995},
month = {July},
}
author = {Geoffrey Gordon},
title = {Online Fitted Reinforcement Learning},
booktitle = {Proceedings of ICML '95 Value Function Approximation in Reinforcement Learning Workshop},
year = {1995},
month = {July},
}