Skip to main content

New answers tagged

0 votes

How to use reinforcement Learning to solve DP with continuous action?

Applying constraints to the outputs of the neural network is hard. Instead apply the constraints to how the distribution is interpreted, either when you sample, or move the decision into (a ...
Neil Slater's user avatar
  • 32.7k
2 votes

Why parameter-based RL methods are not widely used?

what you are talking about are evolutionary strategies, also known as zero order methods They try to approximate: $$ \nabla_\theta \mathbb{E}^{\pi_\theta}[\sum \gamma^t r_t] $$ and they do so by using ...
Alberto's user avatar
  • 2,333
1 vote
Accepted

Actor Critic need to find the goal to have good update and suceed?

I would like to know if im right and if there is method to solve that or if TD method are just useless in this case ? Your assessment is partially correct, but TD learning algorithms have a behaviour ...
Neil Slater's user avatar
  • 32.7k
1 vote

Learning rate and rewards in Deep Reinforcement Learning

I think that you should use quite different approaches when training on environments with sparse rewards compared to training on environments with dense rewards. It is not just the learning rate. With ...
pi-tau's user avatar
  • 825
0 votes

Learning rate and rewards in Deep Reinforcement Learning

If you have sparse binary reward, most of the rewards are $0$, so if you are using DQN, the network quickly learns to always predict $0$ (and the gradient would be 0 since there is no error), and in ...
Alberto's user avatar
  • 2,333
0 votes

Can all RL algorithms learn with discrete state spaces?

Two things in this post need to be corrected. Q-learning is not married to an estimator. For instance, DQN stands for Deep Q network, a type of Q-Learning that utilizes neural networks. On the ...
foreverska's user avatar
  • 1,318
3 votes
Accepted

Why do big policy updates cause performance drop in deep RL?

One prominent reason for preventing large policy updates has to do with the architecture of deep reinforcement learning algorithms: since parameters are shared by all inputs in a neural network, ...
DeepQZero's user avatar
  • 1,494

Top 50 recent answers are included