Hi everyone !
Please use this section to ask your questions and upvote other students' questions, for the in-person Q&A session next week. (by Sunday 12, June)
The questions with the most upvotes will be discussed during the Q&A.
Cheers !
The TAs
I have questions about some of the exercises:
Week 5: Ex.4) BackProp with Gaussian units: why is in the reordering of the backward pass the exponential term and the norm term missing?
Week 8: Ex. 5b) I understand the first part, but for the boundary analysis (to get the weights): can someone maybe explain this again?
Thanks a lot!
Week 5: Ex.4) BackProp with Gaussian units: why is in the reordering of the backward pass the exponential term and the norm term missing?
Week 8: Ex. 5b) I understand the first part, but for the boundary analysis (to get the weights): can someone maybe explain this again?
Thanks a lot!
& one general questions about terminology:
If one says for example "we have a 3 layer MLP", does that mean 1 input layer, 1 hidden layer & 1 output layer, and for example 5 layer MLP would be 1 hidden layer, 3 hidden layer & 1 output layer? So when "amount of layers" is mentioned, input and output layers are always included?
If one says for example "we have a 3 layer MLP", does that mean 1 input layer, 1 hidden layer & 1 output layer, and for example 5 layer MLP would be 1 hidden layer, 3 hidden layer & 1 output layer? So when "amount of layers" is mentioned, input and output layers are always included?
Hello,
I have questions about the Reinforcement Learning with Baseline Algorithm (Lecture 4 slide 104). I don't get why we need a discount factor when we update the parameter w (the factor is multiplied before the delta). The derivation for updating theta is shown in the slides, and I get that part, but I am not sure how to derive this discount factor for updating w.
I also have a question about the Actor-Critic with Eligibility traces. There are multiple versions of it provided in lecture 10 (slide 68, 72, and 87). If it is possible, can you explain the differences and when to use which version?
Thanks a lot!
I have questions about the Reinforcement Learning with Baseline Algorithm (Lecture 4 slide 104). I don't get why we need a discount factor when we update the parameter w (the factor is multiplied before the delta). The derivation for updating theta is shown in the slides, and I get that part, but I am not sure how to derive this discount factor for updating w.
I also have a question about the Actor-Critic with Eligibility traces. There are multiple versions of it provided in lecture 10 (slide 68, 72, and 87). If it is possible, can you explain the differences and when to use which version?
Thanks a lot!
I know the deadline has passed, but hey what does it cost.
For the exam 2021 Part 2, I do not really get how we can give an interpretation of Q(tilde) qualitatively. What would a good answer be?
For the exam 2021 Part 2, I do not really get how we can give an interpretation of Q(tilde) qualitatively. What would a good answer be?
Hi ! Same as the previous question, I know the deadline has passed but I was wondering if it would be possible to go over the forward and backward pass in the third section of the 2021 exam ? Thanks !
