This piece was first published in LinkedIn by Bradley Dice, U-M Ph.D student in Physics and Computational Science.
MICDE Symposium 2018: Computation, A Pillar of Science and a Lens to the Future
High-performance computing (HPC) is becoming an increasingly powerful tool in the hands of scientists, driving new discoveries in physical sciences, life sciences, and social sciences. The development of new (frequently domain-specific) approaches to machine learning and faster, smarter processing of sets of Big Data allows us to explore questions that were previously impossible to study. Yesterday, I presented a poster at the Michigan Institute for Computational Discovery & Engineering (MICDE) annual Symposium and attended a number of talks by researchers working at the intersection of high-performance computing and their domain science. The theme for the symposium was “Computation: A Pillar of Science and a Lens to the Future.”
Collaborative Computational Science with signac
My scientific work, and the work of my colleagues in the Glotzer lab, has been made vastly more efficient through the use of tools for collaborative science, particularly the signac framework. I presented a poster about how the signac framework (composed of open-source Python packages signac, signac-flow, and signac-dashboard) enables scientists to rapidly simulate, model, and analyze data. The name comes from painter Paul Signac, who, along with Georges Seurat, founded the style of pointillism. This neo-impressionist style uses tiny dots of color instead of long brushstrokes, which collectively form a beautiful image when the viewer steps back. This metaphor fits the way that a lot of science works: given only points of data, scientists aim to see the whole picture and tell its story. Since our lab studies materials, our “points” of data fit into a multidimensional parameter space, where quantities like pressure and temperature, or even particles’ shapes, may vary. Using this data, our lab computationally designs novel materials from nanoparticles and studies the physics of complex crystalline structures.
The core signac package, which acts as a database on top of the file system, helps organize and manage scientific data and metadata. Its companion tool signac-flow enables users to quickly define “workflows” that run on supercomputing clusters, determining what operations to perform and submitting the jobs to the cluster for processing. Finally, signac-dashboard (which I develop) provides a web-based data visualization interface that allows users to quickly scan for interesting results and answer scientific questions. These tools include tutorials and documentation, to help users acquaint themselves and get on to doing science as quickly as possible. Importantly, the tools are not specific to materials science. Many scientific fields have similar questions, and the toolkit can easily be applied in fields where exploration or optimization within parameter spaces are common, ranging from fluid mechanics to machine learning.
During the symposium, I learned a lot about how others are using scientific computing in their own work. The symposium speakers came from a wide range of fields, including biology, mathematics, and fluid dynamics. Some of my favorite talks are described below.
The Past: Phylogeny and Uncovering Life’s Origins
High-performance computing is enabling scientists to look in all sorts of directions, including into the past. Stephen Smith, Assistant Professor of Ecology and Evolutionary Biology at the University of Michigan, talked about his lab’s research in detecting evolutionary patterns using genomic data. From the wealth of genetic data that scientists have collected, the Smith lab aims to improve our understanding of the “tree of life”: the overarching phylogenetic tree that can explain the progress of speciation over time. Projects like Open Tree of Life and PHLAWD, an open-source C++ project to process data from the National Center for Biotechnology Information’s GenBank data source, are just two of the ways that open science and big data are informing our understanding of life itself.
The Present: From Algebra to Autonomy
Cleve Moler, the original author of the MATLAB language and chief mathematician, chairman, and cofounder of MathWorks, spoke about his career and how the tools MATLAB has provided for numerical linear algebra (and many other computational tasks) have been important for the development of science and engineering over the last 34 years. MATLAB is taught to STEM students in many undergraduate curricula, and is used widely across industry to simulate and model the behavior of real systems. Features like the Automated System Driving Toolbox are poised to play a role in autonomous vehicles and the difficult computational tasks inherent in their operation.
The Future: Parallel-in-Time Predictions and Meteorology
A significant challenge in weather and climate modeling is that supercomputer architectures are highly parallel, while many simulations of fluids are inherently serial: each timestep must be computed before the next timestep can begin. Beth Wingate, Professor of Mathematics at the University of Exeter and published poet, is developing a powerful approach that may change the way that such models work. Called “parallel-in-time,” it separates the effects of slow dynamics and fast dynamics, enabling parallel architectures to take advantage of longer timesteps and separate the work across many processors.
Conclusions
Computational science is growing rapidly, improving our ability to address the most pressing questions and the mysteries of our world. As new supercomputing resources come online, such as Oak Ridge National Laboratories’ Summit, the promise of exascale computing is coming ever closer to reality. I look forward to what the next year of HPC will bring to our world.