37:00 the size of the filter should be 32 x 64 x 1 x 1. another typo. Also the input tensor shape is disproportion. depth (64) should be longer than height and width (56)
at 35:09, the expression for output in case of stride convolution is (W - K + 2P)/S +1...for W=7, K=3, P = (K-1)/2 = 1 & S=2 we get output as (7 - 3 + 2*1)/2 + 1 = 3 +1 = 4 ...however, the slide shows the output as 3x3 instead of 4x4 at the right hand corner... is it correct..?
Thank you for exellent video! But I have a question here, at 1:05:42, after layer normalization, every sample in x has shape 1xD, while μ has shape Nx1. How do you perform the subtraction x-μ?
I wonder if gamma and beta with 1 x D is a typo? If it should be N x 1? If it is not a typo, doing the subtraction is just using the broadcasting mechanism like in numpy.
Just finished watching the lecture, as per my understanding, X (1 X C X H X W) is the shape of the input vector consumed at once in the algo, and for the calculated means and standard deviations they have mentioned the shape of the output vectors of these parameters in terms of batch size (N X 1 X 1 X 1) as each value uniquely represents each input (1 X C X H X W). It is a late reply but I am replying if someone else would scroll through with similar question to yours!
Because both are linear operators, then you can simply concat them after training (think of them as matrices A and B, in test time you multiply C=A*B and you put that instead of both)
When you dot product 3d image example(3*32*32) with filter(3*5*5) gives a 2d feature map (28*28) just bcoz of the dot product operation between image and filter
In Batch Normalization during Test time at 59:52, what are the averaging equations used to average Mean & Std deviation, sigma ..during the lecture some mention is made of exponential mean of Mean vectors & Sigma vectors...please suggest.
Great lecture for free. Thank you Michigan University and professor Justin.
I have this great package for my university course.❤
Thank you for the great course.
Thank you I found answers to the questions that I have been looking for long time
33:10 stride
53:00 batch normalization
37:00 the size of the filter should be 32 x 64 x 1 x 1. another typo. Also the input tensor shape is disproportion. depth (64) should be longer than height and width (56)
Amazing!
Pd: Although I am sorry for the guy with the coughing attack...........
yeah, kinda disturbed me to concentrate. 2019 it was right before covid striked the world hahah
😷
at 35:09, the expression for output in case of stride convolution is (W - K + 2P)/S +1...for W=7, K=3, P = (K-1)/2 = 1 & S=2 we get output as (7 - 3 + 2*1)/2 + 1 = 3 +1 = 4 ...however, the slide shows the output as 3x3 instead of 4x4 at the right hand corner... is it correct..?
I have the same question.
both are different situations, the calculation is done without padding whereas the formula is written considering padding
@@krishnatibrewal5546 ... thanks a lot, yes you're right..
@@krishnatibrewal5546 thanks.
Thank you for exellent video! But I have a question here, at 1:05:42, after layer normalization, every sample in x has shape 1xD, while μ has shape Nx1. How do you perform the subtraction x-μ?
I wonder if gamma and beta with 1 x D is a typo? If it should be N x 1? If it is not a typo, doing the subtraction is just using the broadcasting mechanism like in numpy.
@@useForwardMax Broadcasting mechanism makes sense. Thank you.
Just finished watching the lecture, as per my understanding, X (1 X C X H X W) is the shape of the input vector consumed at once in the algo, and for the calculated means and standard deviations they have mentioned the shape of the output vectors of these parameters in terms of batch size (N X 1 X 1 X 1) as each value uniquely represents each input (1 X C X H X W).
It is a late reply but I am replying if someone else would scroll through with similar question to yours!
1:01:30 what did he mean by “fusing BN with FC layer or Conv layer”?
You can have conv-pool-batchnorm-relu or fc- bn- relu , batch norm can be induced between any layer of the network
@@krishnatibrewal5546 thanks a lot!
Because both are linear operators, then you can simply concat them after training (think of them as matrices A and B, in test time you multiply C=A*B and you put that instead of both)
Great.
why they don't use batch norm + layer norm together?
Thanks for an excellent video Justin!! I had a quick question on how does the conv. filters change the 3d input into a 2d output
When you dot product 3d image example(3*32*32) with filter(3*5*5) gives a 2d feature map (28*28) just bcoz of the dot product operation between image and filter
In Batch Normalization during Test time at 59:52, what are the averaging equations used to average Mean & Std deviation, sigma ..during the lecture some mention is made of exponential mean of Mean vectors & Sigma vectors...please suggest.
Well don! here is more explanation to normalization: th-cam.com/video/sxEqtjLC0aM/w-d-xo.html&ab_channel=NormalizedNerd
sounds like someone was building duplo the entire lecture
Thomas the tank engine?
ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem ahem .
ahem ahem.
ahe ahe he he HUUUJUMMMMMMMMMMMM
im fucked for this quiz 2 lol