Maybe there is an error in Question 4 Chapter 10

[danli349]

"Consider a CNN that takes in 32×32 grayscale images and has a single convolution layer with three 5×5 convolution filters (without boundary padding)", if without boundary padding, the first convolution layer should be 3@28×28, I don't understand why the answer is 5@32×32?

"(b) How many parameters are in this model?" 5×5×3+3=78
"(d) If there were no constraints, then how many weights would there be in the ordinary feed-forward neural network in (c)?"
If the convolution layer still 32×32, then the parameters should be (32×32+1)×32×32×3 = 3148800, each pixel in the convolution layer should have a bias, I am not sure am I correct?

[danhalligan]

There's definitely some errors here I think. Hard to know what I was thinking at the time, but I think 3 vs 5 layers is just a mis-read of the question from me and 32x32 pixels I think is also an error (I hadn't considered the boundary padding issue). This means my calculation of the number of parameters (part b) is wrong and I agree with you it should be 78.

For part d, I think you're correct, but might need a bit of time to convince myself (with the adjustment for 32 -> 28 pixels from the above error).

I'll try and update when I get chance

[danhalligan]

Thanks @danli349 for another good spot! Hopefully fixed now.