BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Jonathan Eckstein (Rutgers University) DTSTART:20210222T143000Z DTEND:20210222T153000Z DTSTAMP:20260423T035056Z UID:OWOS/37 DESCRIPTION:Title: Pr ogressive Hedging and Asynchronous Projective Hedging for Convex Stochasti c Programming\nby Jonathan Eckstein (Rutgers University) as part of On e World Optimization seminar\n\n\nAbstract\nOperator splitting methods for convex optimization and monotone inclusions have their roots in the solut ion of partial differential equations\, and have since become popular in m achine learning and image processing applications. Their application to " operations-research-style" optimization problems has been somewhat limited .\n\nA notable exception is their application to stochastic programming. In a paper published in 1991\, Rockafellar and Wets proposed the progressi ve hedging (PH) algorithm to solve large-scale convex stochastic programmi ng problems. Although they proved the convergence of the method from firs t principles\, it was already known to them that PH was an operator splitt ing method.\n\nThis talk will present a framework for convex stochastic pr ogramming and show that applying the ADMM (and thus Douglas-Rachford split ting) to it yields the PH algorithm. The equivalence of PH to ADMM has lo ng been known but not explicitly published.\n\nNext\, the talk will apply the projective splitting framework of Combettes and Eckstein to the same f ormulation\, yielding a method which is similar to PH but can be implement ed in a partially aynchronous manner. We call this method "asynchronous p rojective hedging" (APH). Unlike most decomposition methods\, it does not need to solve every subproblem at every iteration\; instead\, each iterati on may solve just a single subproblem or a small subset of the available s ubproblems.\n\nFinally\, the talk will describe work integrating the APH a lgorithm into mpi-sppy\, a Python package for modeling and distributed par allel solution of stochastic programming problems. Mpi-sppy uses the Pyomo Python-based optimization modeling sytem. Our experience includes using 8\,000 processor cores to solve a test problem instance with 1\,000\,000 s cenarios.\n\nThis talk presents joint research with Jean-Paul Watson (Lawr ence Livermore National Laboratory\, USA)\, and David Woodruff (University of California\, Davis).\n\nThe address and password of the zoom room of t he seminar are sent by e-mail on the mailinglist of the seminar one day be fore each talk\n LOCATION:https://researchseminars.org/talk/OWOS/37/ END:VEVENT END:VCALENDAR