For the problem 7 of the exam 2018, to make sure that I understood well weight sharing, we have delta_L/delta_w2 = 2 right?
And for the problem 9, I didn't understood well the answer can you explain it in another way please? :/
Yes your correct about your statement with regards to the weight sharing for delta_w2. The reason why we dont have enough information in exercise 9 is because we have changed the whole system so the derivatives in given in the begining of the question are no longer valid and needs to be recalculated and we dont have enough information to do that
Ok thank you so much for your answer!
So, when we change the whole system as in problem 8 or 9, we cannot say anything without recalculating the derivatives, except in the special case of problem 8 when w1 is not connected anymore to the output. Correct? :)
Hi,
could you explain why deltaL/delta_w2=2 when we have weight sharing with w3?
I guess by deduction that it is the sum ( deltaL/deltaw2 +deltaL/deltaw3 ), but i don't get the link with the definition of weight sharing. Isn't it supposed to only mean that w2=w3?
thank you so much for your help!
Pb 7 and 9 exam 2018
Hello,
For the problem 7 of the exam 2018, to make sure that I understood well weight sharing, we have delta_L/delta_w2 = 2 right?
And for the problem 9, I didn't understood well the answer can you explain it in another way please? :/
Thank you so much for your help!
Yes your correct about your statement with regards to the weight sharing for delta_w2. The reason why we dont have enough information in exercise 9 is because we have changed the whole system so the derivatives in given in the begining of the question are no longer valid and needs to be recalculated and we dont have enough information to do that
Ok thank you so much for your answer!
So, when we change the whole system as in problem 8 or 9, we cannot say anything without recalculating the derivatives, except in the special case of problem 8 when w1 is not connected anymore to the output. Correct? :)
Yes exactly!
1
Hi,
could you explain why deltaL/delta_w2=2 when we have weight sharing with w3?
I guess by deduction that it is the sum ( deltaL/deltaw2 +deltaL/deltaw3 ), but i don't get the link with the definition of weight sharing. Isn't it supposed to only mean that w2=w3?
thank you so much for your help!
Yes it is in the lecture about convolutional networks. Yes when weights are shared there are the same there is no 2 weights but only 1.
5
Add comment