Session

DescriptionScientists in all domains increasingly rely on high-throughput applications that combine multiple components into increasingly complex multi-modal workflows executing on heterogeneous systems. The complexities of those applications and their workflows hinder the scientists’ ability to generate results in a robust way. Robust science should assure performance scalability in space and time; trust in technology, people, and infrastructures; and reproducibility or confirmable research in high-throughput applications. Today high-throughput applications are far from achieving these goals.

Through the presentations of a set of comprehensive, cross-disciplinary studies of high-throughput applications for scientific discovery we outline challenges to reach robust science with hardware and systems all the way to policies and practices. We bring together a cross-disciplinary community including computer and data scientists, physicists, natural scientists, and molecular dynamics scientists discussing practices and procedures to help define, design, implement, and use a set of solutions for robust science. We take important steps to define a roadmap that enables high-throughput applications to withstand and overcome adverse conditions such as heterogeneous, unreliable architectures at all scales including extreme scale, lack of rigorous testing under uncertainties, unexplainable algorithms (e.g., in machine learning), and black-box methods.