CBC Talk on HPC Considerations for Robust, Reliable Optimization Codes - February 17, 2015
Total number of participants: 8
Total number of guests outside of CBC: 1
Number of different nationalities represented: 5
Total number of speakers: 1
Total number of talks: 1
Computational tools for optimal design and parameter estimation require us to balance two different challenging areas. On one hand, we must implement large-scale, parallel, differential equation (PDE) solvers. On the other, we must integrate these PDE solvers into a robust, reliable optimization framework. Focusing on the latter, we observe that optimization solvers must account for several considerations unique to these kinds of problems. For example, explicit time integrators, used in PDE solvers, have stability limits that the optimization solver may exceed, so our codes must be robust towards unstable solves. Large parallel clusters crash and required maintenance, which disrupt long running computations. Therefore, our codes must be reliable to hardware faults. How we parallelize our solves has changed over time as we migrate from traditional clusters to GPUs to new paradigms required for exascale computing. In response, our optimization codes must be adaptable to changing definitions of parallelism.
In the following presentation we give an overview of these challenges and highlight the features, properties, and attributes required by optimization solvers to give robust, reliable results.
Center for Biomedical Computing (CBC) aims to develop and apply novel simulation technologies to reach new understanding of complex physical processes affecting human health. We target selected medical problems where insight from mathematical modeling can contribute to changing clinical practice.