4 ms·
How do they sort the dimensions so the most salient are at the front and thus retained ?
by deepnet 3y ago
How do they sort the dimensions so the most salient are at the front and thus retained ?
- yorwba 3y agoThey declare by fiat that retaining only the first few dimensions should lead to low classification error and construct the training loss accordingly. (Equation 1 on page 4 of the PDF.)