New SISC Section on Scientific Machine Learning

By Hans De Sterck

Beginning with Volume 46, Issue 1, the SIAM Journal on Scientific Computing (SISC) will utilize a new section structure that reflects the evolution of the scientific computing field. Going forward, the journal will feature the following three sections:

• Numerical Algorithms for Scientific Computing: This is SISC’s "classic" section, publishing the latest advances in numerical methods that have broad relevance for computationally challenging real-world problems in science, engineering, and other areas of application.
• Software, High-Performance Computing (HPC), and Computational Science and Engineering (CSE): This section publishes papers specifically addressing novel algorithmic contributions in mathematical software or in high-performance computing (HPC), or numerical methods developed for a specific problem in computational science and engineering (CSE).
• Machine Learning Methods for Scientific Computing: This new section publishes papers that develop novel machine learning methods for solving mathematical modeling problems in science and engineering. Papers using scientific computing techniques for developing new machine learning methods with applications in other areas than science and engineering are also of interest if they are written to appeal to a broad scientific computing audience.

Why are we adding this new section?

SISC’s new section, ‘Machine Learning Methods for Scientific Computing’, reflects a major development in scientific computing over the past decade: the emerging use of data-driven and machine learning methods to solve problems in science and engineering in entirely new ways, addressing problem classes that were often out of reach for traditional methods in terms of problem complexity and dimensionality. Scientific machine learning is quickly developing into a major area of scientific computing. For example, methods based on deep learning paradigms are now used extensively as numerical methods to solve challenging real-world problems in many areas of application.

What types of papers should authors submit?

SISC’s new scientific machine learning section affirms SISC as a natural home for strong and impactful papers in this area, recognizing the far-reaching promise these types of numerical methods hold for the future of scientific computing. Papers in this section should present new machine learning methods that form accurate, efficient, robust, and scalable algorithms, with a mathematically precise description of the new techniques. Papers that use scientific computing techniques to tackle the tremendous computational challenges in machine learning are also of interest to this section. The new algorithms presented in this section must be motivated by theory or strong heuristics and must demonstrate strong computational performance using carefully designed numerical experiments, in comparison with existing state-of-the-art algorithms. What sets papers in this section apart from other venues in the machine learning literature is the thoroughness and rigor of SISC’s review process with a focus on reproducibility. These high standards ensure that publications in this section realize lasting impact and help shape the emerging field of scientific machine learning.

Software, HPC, and CSE were introduced as new focus areas for SISC in 2012 when SISC’s section structure was initiated. The integrated section on “Software, HPC, and CSE” reaffirms SISC’s interest in these crucial aspects of contemporary scientific computing. SISC welcomes papers that present algorithmic or technological innovations in mathematical software, for example, in areas that include software for numerical methods for PDEs, numerical linear and nonlinear algebra, mesh generation, and automatic differentiation. HPC papers typically focus on topics such as parallel algorithms, combinatorial scientific computing, finite-precision arithmetic, quantum numerical algorithms, or new computer architectures. CSE papers target a specific challenging problem in computational science or engineering. They may advance fundamental algorithms, or their innovation may consist of the creativity needed to synthesize a computational solution from the right algorithmic building blocks, solving problems in new and superior ways.

What changes should be expected?

Overall, this new section structure for SISC will change neither the purpose nor the quality of the journal. SISC publishes major advances in numerical method development that are of general relevance across broad application areas. SISC dates to 1980, when the journal was born as the SIAM Journal on Scientific and Statistical Computing. Its name was changed to the SIAM Journal on Scientific Computing in 1993. The traditional focus areas of the journal include numerical methods for PDEs and numerical linear algebra with applications in science and engineering. However, since the inception of the journal in the 1980s, the scope of SISC and the field of ‘scientific computing methods’ that the journal defines have expanded to encompass any method for numerical computing that is relevant for any computationally challenging real-world problem, with applications that go far beyond science and engineering. SISC is SIAM’s general-purpose numerical methods journal, and it now plays a key role in broad areas of the global research community, given the ever-increasing importance of numerical methods for ever-expanding areas of knowledge. When machine learning methods are used as numerical methods to solve challenging real-world problems in science, engineering and beyond, it is natural to incorporate scientific machine learning into SISC as a new emerging branch of scientific computing.

We look forward to the success of this new section! Access SISC and submit your relevant papers.

Hans De Sterck is a professor of applied mathematics at the University of Waterloo. He is the editor-in-chief of the SIAM Journal on Scientific Computing and director of Waterloo’s Centre for Computational Mathematics in Industry and Commerce. His research focuses on numerical methods for computational science and data science.