2-27

• 0:25:52: Q1 What is the output size of 1x1 conv with 128 filters? ANS: 28x28x128

• 0:27:54: Q2 how about the output of 3x3 conv, what is the output size? ANS: 28x28x192

• 0:40:32: Q3 What are the circled network branches for? ANS: training

• 0:49:43: Q4 Why do we want to avoid fully connected layer? ANS: lots of parameters and inflexible input dimension

• 1:01:31: Q5 why do you think residual network work so well? ANS: skip connection