Argonne National Laboratory

Randomized Heuristics for Exploiting Jacobian Scarcity

TitleRandomized Heuristics for Exploiting Jacobian Scarcity
Publication TypeJournal Article
Year of Publication2012
AuthorsLyons, A, Safro, I, Utke, J
JournalOptimization Methods & Software
Start Page311-322
Date Published04/2012
Other NumbersANL/MCS-P1716-0110

We describe a code transformation technique that, given code for a vector function F, produces code suitable for computing collections of Jacobian-vector products F\'(x)[x dot] or Jacobian transpose-vector products F\'(x)[sup T][y bar]. Exploitation of scarcity � a measure of the degrees of freedom in the Jacobian matrix � means solving a combinatorial optimization problem that is believed to be hard. Our heuristics transform the computational graph for F, producing, in the form of a transformed graph G\', a representation of the Jacobian F\'(x) that is both concise and suitable for the evaluation of large collections of Jacobian-vector products or Jacobian-transpose-vector products. Our heuristics are randomized in nature and compare favorably in all cases with the best known heuristics.