Join the Chemical Engineering Department for a seminar with Carl Laird of Sandia National Lab. Hosted by CMU ChemE Professor Chrysanthos Gounaris.
The passcode for this meeting is 161546.
Title: Accelerating Impact on Global Challenges with High-Performance Optimization
The COVID-19 pandemic has clearly exposed significant challenges in effective mitigation of emerging infectious diseases. Given the impact of spatio-temporal heterogeneities and inter-patch mobility, there is a need for national and global scale models with high-spatial resolution. Unfortunately, traditional sampling-based techniques are not tractable for inference on these models. As part of a COVID-19 rapid response project, we have developed large-scale optimization techniques that can efficiently estimate county-level transmission parameter dynamics using a fully-coupled, national-scale model. Laird will discuss the approach in our open-source inference package and our ongoing collaboration to quantify the impact of mitigation strategies in the US.
Presidential Policy Directive 21 identified 16 critical infrastructure sectors, including select manufacturing, chemical, water, and electrical systems. Safe, reliable performance of these critical infrastructures is threatened by extreme weather events, deterioration due to aging, increased interdependence, and climate change. Challenging design, retrofit, and operations problems in critical infrastructure are represented as mixed-integer nonlinear programming problems that cannot be addressed with off-the-shelf solvers. Laird will briefly present on the success of tailored approaches for these problems and describe our current work investigating the use of machine learning models to replace challenging, large-scale constraints.
These important applications push the boundaries of computational performance, and we need approaches to solve optimization problems faster while addressing larger-scale, increasingly integrated systems. Hardware performance improvements in scientific computing applications are being realized through multi-core processors, distributed clusters, and specialized accelerator architectures. Laird will briefly discuss our work on scalable parallel algorithms for distributed clusters and graphics processing units, and he will show that rapid development of tailored approaches is possible with high-level languages.