luca
Posts:
11
Registered:
5/19/10


Re: Matrix multiplication by its transpose
Posted:
Jun 9, 2010 9:55 AM


On 9 Giu, 15:15, Brian Borchers <borchers.br...@gmail.com> wrote: > On Jun 9, 5:40 am, luca <luca.frammol...@gmail.com> wrote: > > > > > > > Hi, > > > i have fear of asking this question, because it could be stupid, but i > > will try it nevertheless: > > > suppose i have a matrix A N x 8, where N is of order 10^310^5 and a > > vector v of size Nx1 (all are real elements). > > > To compute the element (i, j) of A i need to do some computation > > (derivative of image pixels, some multiplications and so on). After > > computing A, i need to compute its pseudoinverse: pinvA = (A^T * > > A)^1 * A^T (where A^T is the transpose of A and (..)^1 is the > > inversione of the quantity between parenthesis) and finally i need to > > compute the multiplication of the pesudoinverse by v: > > > pInvA * v > > have you considered using the singular value decomposition to find the > pseudoinverse of A? This would save you the trouble of multiplying > out A'*A and avoid the illconditioning of the A'*A matrix. > > Yes, 1.0e11 is a big condition number you want to avoid working with > A'*A if possible. Nascondi testo citato > >  Mostra testo citato 
Hi, mmm, don't know, the SVD of A should be too heavy from a computational point of view (i am writing a realtime application)...

