What is the difference between adaptive and average pooling? I couldn’t find this in Chapter 14.
Is there a full implementation of ResNet as described in Chapter 14 available? I looked in the GitHub repo but couldn’t find one.
Hi all.
I was playing with the convolutional neural network with different kernels on MNIST digit data. I changed kernels for Conv2d layers to detect the top and right edges. Then, I combined the outputs of those two kernels for a new output. I wrote a blog on it if anybody wants to take a look and try it out.
Have fun.
1 Like
Principle Component Analysis is topic in linear algebra. Basically, you are attempting to create an alternative basis to capture the maximum amount of variability with a minimum of dimensions (fewest basis vectors). For example, the vector from (0,0,0) to (1, 2,3) is a straight line being represented in three dimensional space. If you orient the frame to have the first basis on the (1,2,3) vector, you could capture the variability with one dimension rather than three, requiring only one value to describe the vector.
The main point is dimension reduction, which is why PCA was used to transform the movie latent variables from five dimensions to two for plotting.