flâneur

High Dimensions.pdf

people.eecs.berkeley.edu · 6,559 words · saved by 1 readers

N/A

134 Jonathan Richard Shewchuk 22 The Pseudoinverse; Better Generalization for Neural Nets THE PSEUDOINVERSE AND THE SVD [We’re done with unsupervised learning. For the rest of the semester, we go back to supervised learning.] [The singular value decomposition can give us insight into the pseudoinverse and its use in least-squares linear regression. If you attended Discussion Section 6, you worked through an explanation of this, but now that I’ve introduced the compact SVD in Lecture 21, I’d like to summarize it.] Let X be any n ⇥ d matrix. Let…

related reading