Summer Student Symposium

Event Sponsor: 
LANS Informal Seminar Series
Start Date: 
Aug 8 2008 (All day)
Building 221, Conference Room tba
Argonne National Laboratory
Four Students
Speaker(s) Title: 
MCS Summer Students
Sven Leyffer

Summer Student Symposium, a series of talks from our best 2008 summer students.

* Kyle P Schmitt: Two Creative and Expeditious Variations of Traditional Gaussian Processes
* Nawab Ali: Rethinking I/O in High-Performance Computing Environments
As the types of problems we solve in high-performance computing and other areas become more complex, the amount of data generated and used is growing at a rapid rate. Today many terabytes of data are common; tomorrow petabytes of data will be the norm. One of the challenges in high-performance computing is to provide users with reliable data access in a distributed, heterogeneous environment. In this talk, I will review the existing I/O paradigms in high-performance computing environments and suggest better alternatives across both local and wide-area networks.
* Sean Farley: Enabling PETSc preconditioners in BOUT code for Edge Plasma Modeling
Developing an interface between the BOUT edge turbulence application (developed at Lawrence Livermore National Laboratory) and the timestepping and preconditioned Newton-Krylov methods in the PETSc library.
* Heather Cole-Mullen: Implementing Automatic Differentiation Tools in the NEOS Server for Optimization
Creating a new infrastructure for Automatic Differentiation-enabled solvers in the Network Enabled Optimization Solvers Server, generalizing and augmenting the implementation of AD in NEOS, and developing a method for the comparison of available AD tools on a number of NEOS solvers.
* Joseph Reed: Benchmarking Pattern-Search and Nelder-Mead Optimization Algorithms
Pattern-Search and Nelder-Mead algorithms belong to the derivative free class of methods. In order to test which derivative free algorithm is more effective we run each on problems in which the objective function is potentially not convex, not smooth, and expensive to evaluate. The questions we are seeking to answer include, amongst other things, include how sensitive each algorithm is to the starting point.
* Yuchen Wu: Reduced Space Quasi Newton Methods for PDE Constrained Optimization
This talk is to introduce a reduced space quasi newton method for solving PDE constrained optimization. PDE constrained optimization is widely applicable to various fields of science and engineering. However, in most cases these problems are exceedingly hard to solve, due to large dimensions and difficulty with evaluating derivatives. This forces algorithms to i) bring down the number of linear solves and avoid factorization; ii) avoid use of second order derivatives. In the proposed method we attempt to address the two requirements. Numerical results are shown to compare reduced space method with full space method; furthermore, three variants of the reduced space methods are compared.
* Mustafa Kilinc: ASTROS: Active-Set Trust-Region Optimiation Solvers
We describe a new open-source framework for solving nonlinear optimization problems. The solvers are based on a range of active-set step computation, and implement a range of different globalization strategies such as penalty functions, filters, and tubes. We present some preliminary numerical experience.

Miscellaneous Information: