Title of talk:
Computational Progress on Mathematical Optimization Solvers
Abstract of talk:
We describe some recent advances in developing general-purpose numerical solvers for linear and semidefinite programming (LP/SDP), where Don Godfarb was one of the pioneers on the field. The developments are bsed on methods/theories such as Online LP Variable-Classification, LP Cross-Over, the First-Order Hype-Gradient, ADMM, a low-rank SDP theorem similar to the Johnson-Lindenstrauss embedding lemma, and/or careful implementations on GPU machines. These techniques have increased the average solution speed by over 3x in the past three years on a set of benchmark problems. For certain problem types, the speedup is more than 10000x, and problems that have taken days to solve or never been solved before are now solved in seconds or minutes to high accuracy.
