Hello,
I have recently published an algorithm that computes the Moore-Penrose
pseudoinverse, and does so in exact arithmetic. It accomplishes this with 2mn^2
multiplications (dimensions with m>n), thus is comparable to a single matrix
multiply. I believe your function makes use of singular value decomposition.
Mine does not achieve the singular values, which is why it works quicker for
just the pseudoinverse.
Here is the (thus far only) link to my algorithm:
http://math.stackexchange.com/a/317053/43193
I believe that this could enhance your wonderful software, how may I help this
happen? Do I need to submit publication into a journal, or maybe write some
code? (I have my code in python currently, but the algorithm is quite simple
really and could easily be ported-- at least it would be easy for me to do.)
-Adam
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.eecs.utk.edu/mailman/private/lapack/attachments/20130302/db3bdb3a/attachment.html
|