Workshop on Scientific Machine Learning Brings Researchers Across Academia, Industry Together

What are the greatest societal problems that science seeks to solve?  

The development of personalized medicine, perhaps. Building sustainable urban infrastructure. Restoring or generating access to clean air and water. Developing clean energy. Understanding the impacts and consequences of climate change around the world. These problems are vast and disparate, but the difficulties intrinsic in solving them are based on a common denominator: data is often difficult (or expensive) to acquire, dynamics are highly complex, quantifying uncertainty is critical, and decisions often have life-altering consequences. 

Scientific Machine Learning combines perspectives across disciplines to craft machine learning methods that are able to make interpretable predictions that seek to provide insight into how to effectively target these issues. The Oden Institute for Computational Engineering and Sciences core faculty members Stella Offner and Tan Bui-Thanh brought together over 100 researchers and leaders from across the scientific machine learning community for the 2nd Annual Workshop on Scientific and Machine Learning. The event, held Oct. 2-4 at the Oden Institute, provided an opportunity to share research, foster collaboration, and establish central challenges and research directions.

Featured talks were given by university researchers, as well as national lab and industry scientists. The conference began with a talk that can be summarily described as out of this world, and featured Flatiron Institute researcher Francisco Fillaescusa’s discussion of the opportunities and challenges surrounding the use of Machine Learning for Cosmology. 

Day One featured a career panel moderated by Tan Bui-Thanh, with Stella Offner, Tim Wildey, and Leonardo Zepeda-Núñez as panelists, who provided advice surrounding career development and opportunities within academia, national laboratory research, and industry within the scientific machine learning space. Offner highlighted that the ability to work with students was “the most valuable part” of a career in academia, and noted that learning how to “write as well as possible” would serve students well regardless of the opportunities they aim to pursue. 

Wildey, a former Oden Institute postdoctoral researcher and current researcher at Sandia National Laboratories noted that the freedom to choose projects he was passionate about was the biggest benefit of a career in the national labs, but noted difficulties with interfacing with bureaucracy and funding competition. Zepeda-Núñez emphasized the rapid timeline he interfaces with as an industry researcher at Google, and gave students valuable advice surrounding recruiting. “Apply a lot, and remember the law of large numbers,” he shared, adding that industry careers “come with significant stress and job insecurity, but they pay off.”

Day Two of the conference featured talks from many University of Texas at Austin and Oden Institute researchers, including Willerson Center for Cardiovascular Modeling and Simulation Director Michael Sacks, who discussed the development of “rapid, accurate, and detailed” models of the heart, and highlighted that machine learning methods have the potential to “provide simulation results within 2-3 seconds, suffering no loss of accuracy.” 

Nicole Aretz, a postdoctoral researcher with the Willcox Research Group at the Oden Institute, presented her research, which focuses on how uncertainty quantification contributes to making accurate estimates of sea-level rises as a consequence of glacier melting. “The paper was written in a manner accessible to glaciologists, in order to lower the barrier to entry to scientific machine learning spaces,” Aretz said. Her talk was followed by lunch and a poster session, featuring presentations across many disciplines.

Day Three featured a talk from Qiang Sun, a researcher from the University of Chicago, who discussed the role that AI models can play in predicting Gray Swan events - predictable but unlikely occurrences with severe impact. In the aftermath of Hurricane Helene, improving models’ speed and accuracy is vital. Sun discussed the challenges associated with using AI models to predict these events, as well as the workarounds necessary to ensure model validity. The conference concluded with presentations from Luke McLennan and Hai Nguyen,  graduate students with the Computational Visualization Center and the Probabilistic and High Order Inference, Computation, Estimation, and Simulation Group, both representing the Oden Institute, respectively. 

 

As researchers gathered in discussion at the conference’s close, it was clear that its intention was realized. In bringing experts and students within the field of Scientific Machine Learning together, connections were built, ideas were developed, and our understanding of the world’s most complex problems broadened.