After Equation 2.67, the new paragraph says that since we're using the same matrix D to decode all the points, we can no longer consider the points in isolation. Then the author uses the Frobenius norm of the matrix instead of the L2 norm used before. So, why is that?

Also, what are the limits of summation in the norm and how are we defining the transformation on a scalar, ie r(x⁽ⁱ⁾)_j

A discussion for people to introduce themselves. Share your level of expertise, what you're looking for in this community, how you got started in deep learning, or anything else you'd like!

A line in first paragraph says "Defining P(y=1|x)=sigmoid(z) is equivalent to defining P(y=1|x) = softmax(z)1 with two-dimensional z and z1=0."

Could anybody help me understand this line?