By Gitta Kutyniok, Ali Pinar, and Joel A. Tropp
The 21st century has been called the “age of data science,” since people now collect massive volumes of data on just about everything. This data facilitates new approaches to problems in science, technology, engineering, and society.
Mathematics is a necessary component to the development of principled and efficient methodologies for data analysis. For example, machine learning algorithms process data to make scientific discoveries; run critical infrastructure operations; and support vital decisions in a wide variety of fields, including healthcare, loan approval, and even parole. To deploy machine learning tools in high-consequence circumstances, researchers must formulate rigorous performance standards and design algorithms that meet these standards. Mathematics is the language they use to express the resulting guarantees.
Over the years, SIAM has taken a leading role in the advancement of data science. In 2019, the SIAM Journal on the Mathematics of Data Science published its first issue. And this year saw the first installment of a new biennial conference series: the SIAM Conference on the Mathematics of Data Science (MDS). A new activity group on data science (SIAG/DATA) has also been approved.
The goal of the MDS conference series is to convene an interdisciplinary community of researchers who are building rigorous foundations for data science and making principled applications to science, engineering, and beyond.
MDS20 was originally scheduled for May 2020 in Cincinnati, Ohio. The call for papers generated 156 minisymposium proposals, 148 contributed talks, and 39 contributed posters. As co-chairs for MDS20, we recruited seven founders of the field of mathematical data science to present plenary talks. Carola-Bibiane Schönlieb (University of Cambridge) and Ozan Öktem (KTH Royal Institute of Technology) organized a minitutorial on “Deep Learning for Inverse Problems and Partial Differential Equations.” We anticipated over 1,000 attendees, which is unprecedented for the first installation of a SIAM conference.
However, as the threat of the ongoing COVID-19 pandemic became clear in early 2020, SIAM leadership chose to cancel the in-person meeting in the interest of safety. Instead, we reimagined MDS20 as an online event — the first virtual conference in SIAM’s history.
After much last-minute planning, MDS20 took place online from May 4 through June 30, 2020. All seven plenary speakers graciously agreed to give virtual talks over the two-month period. Their talks covered a range of fundamental topics in data science:
- Michael Jordan (University of California, Berkeley) discussed the decision-making side of machine learning and focused on computational, inferential, and economic perspectives.
- Cynthia Dwork (Harvard University) introduced the concept of differential privacy, a mathematical approach that prevents identification of individuals’ data.
- Jennifer Chayes (Berkeley) spoke about modeling and inference for large, sparse graphs using the concept of a graphon (or graph limit).
- Andrea Bertozzi (University of California, Los Angeles) presented methods for the analysis of graph data with tools from numerical partial differential equations, variational analysis, and imaging science.
- David Donoho (Stanford University) surveyed recent literature on scaling up COVID-19 testing using pooled samples.
- Yann LeCun (Facebook and New York University) addressed the nexus between applied mathematics and deep learning, with a focus on open challenges.
- Yurii Nesterov (University of Louvain) described higher-order optimization algorithms that are designed for large-scale machine learning applications.
Each plenary talk attracted between 200 and 500 live viewers. In addition, the minitutorial took place as a six-part virtual event in late May. Speakers included Weinan E (Princeton University), Eldad Haber (University of British Columbia), Ozan Öktem (KTH Royal Institute of Technology), Christoph Reisinger (University of Oxford), Rebecca Willett (University of Chicago), and Lexing Ying (Stanford). Each tutorial talk drew roughly 200 synchronous views. Over the last several months, both the minitutorials and the invited talks have amassed thousands of additional views.
Organizers also held 63 minisymposia throughout the course of the conference. Altogether, these presentations have attracted thousands of synchronous and asynchronous participants. Many of the talks, including the invited lectures, are available on SIAM’s YouTube channel.
Unfortunately, some parts of the conference had to be postponed, including the career fair and a business meeting about SIAG/DATA. Other parts of the meeting, including the contributed talks, became untenable.
As MDS20 co-chairs, we can offer several positive conclusions from our experiences planning an online meeting. Virtual conferences reduce the costs and limitations (monetary, temporal, and carbon) associated with travel. Content is also more widely available—anyone with an internet connection can access the presentations at any time—which may increase participation from early-career scientists and researchers with limited resources or scheduling conflicts.
On the other hand, the logistics of online meetings still require further refinement. For example, it is impossible to accommodate every time zone. We also found that organizing and advertising the event was more challenging, especially because the meeting spanned two months. Subsequent virtual SIAM conferences—including the SIAM Conference on the Life Sciences, the Second Joint SIAM/CAIMS Annual Meeting, the SIAM Conference on Imaging Science, and the SIAM Conference on Mathematics of Planet Earth—have taken place over two- and three-week increments.
In addition, online conferences offer limited occasions for social engagement, which may be detrimental for scientists who wish to expand their research networks. The virtual format also reduces opportunities for serendipitous inspiration. We must not underestimate the importance of personal interaction in the sciences.
Despite the unexpected challenges, we were glad for the chance to offer MDS20 as an online event and maintain some of the meeting’s integrity. We extend our thanks for the dedicated work of the organizing committee, which consisted of Barbara Engelhardt (Princeton), Mark Girolami (Cambridge), Ashish Goel (Stanford), Mason Porter (University of California, Los Angeles), Bernhard Schölkopf (Max Planck Institute for Intelligent Systems), Carola-Bibiane Schönlieb (Cambridge), René Vidal (Johns Hopkins University), Karen Willcox (University of Texas at Austin), Stephen Wright (University of Wisconsin-Madison), and Tong Zhang (Hong Kong University of Science and Technology). With the help of several collaborators, Porter put together a guide to managing an online minisymposium based on his experiences with MDS20.
The SIAM staff also played an invaluable role in managing the physical conference and helping us pivot to the virtual platform. In particular, we would like to thank executive director Jim Crowley; Richard Moore, the director of Programs and Services; and Connie Young and Eva Donnelly of the conferences department.
Plans are already underway for the next MDS conference, which is scheduled to take place as an in-person event in the spring of 2022; the exact dates and location are yet to be determined. We hope to see you there!
Gitta Kutyniok is Einstein Professor of Mathematics at the Technische Universität Berlin. Her research concerns applied harmonic analysis, artificial intelligence, compressed sensing, data science, imaging sciences, and inverse problems. She was chair of the SIAM Activity Group on Imaging Science from 2018-2019 and was elected as a SIAM Fellow in 2019. Ali Pinar is a Distinguished Member of Technical Staff at Sandia National Laboratories in Livermore, Calif. His research concerns network science, cyber analytics, and cyber experimentation. Joel A. Tropp is Steele Family Professor of Applied and Computational Mathematics at the California Institute of Technology. His research concerns data science, random matrix theory, and numerical analysis. He was elected as a SIAM Fellow in 2019.