2 Comments
User's avatar
Adam Ginensky's avatar

"For example, we know precisely how long our algorithms take to multiply and factor matrices." . At least in the case of matrix multiplication the technically correct statement is we have an upper bound on how long it takes to multiply matrices. Current best is O(n^2.37) and I believe that O(n^2) is still believed to be the optimat number (perhaps n^{2 +\epsilon} is correct)

Expand full comment
Ben Recht's avatar

I chose my words carefully there. I didn't say "we know what the optimal algorithm is for matrix factorization." Because there, you are right. We don't know the answer. Instead, I wrote "we know precisely how long our algorithms take to multiply." This refers to the deterministic algorithms implemented in BLAS and LAPACK.

Expand full comment