21 search results - page 2 / 5 » Rates of Convergence of Performance Gradient Estimates Using... |
NIPS
2008
13 years 7 months ago
2008
Policy gradient (PG) reinforcement learning algorithms have strong (local) convergence guarantees, but their learning performance is typically limited by a large variance in the e...
ICML
2003
IEEE
14 years 6 months ago
2003
IEEE
In Reinforcement Learning (RL) there has been some experimental evidence that the residual gradient algorithm converges slower than the TD(0) algorithm. In this paper, we use the ...
ECAI
2008
Springer
13 years 7 months ago
2008
Springer
In this paper, we propose a policy gradient reinforcement learning algorithm to address transition-independent Dec-POMDPs. This approach aims at implicitly exploiting the locality...
NIPS
2001
13 years 7 months ago
2001
Policy gradient methods for reinforcement learning avoid some of the undesirable properties of the value function approaches, such as policy degradation (Baxter and Bartlett, 2001...
NIPS
2007
13 years 7 months ago
2007
We present four new reinforcement learning algorithms based on actor-critic and natural-gradient ideas, and provide their convergence proofs. Actor-critic reinforcement learning m...