Iteration grain sized asynchronous parallel algorithms in optimization

Abstract

A general framework for large-grained parallelization of optimization methods is presented, together with a proof of convergence, For comparison Bertsekas' and Tsitsiklis method is recalled: both of these algorithms have iteration grain size, but the presented new method does not require the independent evaluation of the components. Variants of the new method and respective merits and disadvantages are discussed

Keywords:

• Large-grained parallelization
• asynchronous parallel computation
• optimization techniques
• MIMD computers

^∗ [email protected]

^∗ [email protected]

Notes

^∗ [email protected]