How can we speed up the DQN training process?
2. Question 2 How can we speed up the DQN training process? 1 point Use a smaller model. Do not use pooling. Stack more layers. Use stacked state and…
-
-
How can we help an agent to adapt to different scales of a reward signal?
4. Question 4 How can we help an agent to adapt to different scales of a reward signal? 1 point Give the agent more training samples with higher learning…
-
DQN? Why do we need to stack four last frames while playing Atari games?
1. Question 1 DQN 1 / 1 point Neither off-policy nor on-policy method Is based on an off-policy method. Is based on an on-policy method. Yes, it is…
-
What is true about planning in RL?
1. Question 1 What is true about planning in RL? 1 point Planning allows to compute (contrast with learn)the best possible action. For planning, we do not need…
-
What are different types of planning?
3. Question 3 What are different types of planning? 1 point Background planning starts after an agent’s transition into a new state; it is used to select an optimal…
-
What are the differences between model-free and model-based settings?
2. Question 2 What are the differences between model-free and model-based settings? 1 point In a model-based setting, we know nothing about environment dynamics. An agent is learning by…
-
What does the reward discounting means for an agent?
3. Question 3 What does the reward discounting means for an agent? 1 / 1 point It reduces the variance of the return estimator by decreasing the contribution of distant rewards. Intuitively, think…
-
What of the following may complicate optimization in RL?
2. Question 2 What of the following may complicate optimization in RL? 1 / 1 point Negative feedback loops and careless reward normalization See the lecture for illustration of the…
-
Which of these are correct ways to alter the reward function?
1. Question 1 Which of these are correct ways to alter the reward function? Note: by “correct” we mean that it does not change the optimal policy. 1 / 1 point Reshape…