Data Reduction for Science

The Data Reduction for Science grant is focused on the field of scientific data management, primarily in the reduction and efficiency of data. As scientific research and experiments generate substantial data, researchers face complications in storing, analyzing, and archiving the data in their raw form. To manage these volumes, many researchers have incorporated data reduction techniques like compression, filtering, and feature extraction. But even after reduction, data handling remains a concern, pushing the need for more innovative data reduction methods.

The Integrated Research Infrastructure program recognizes this challenge and encourages the incorporation of mathematical rigor in data reduction. The goal is to retain relevant scientific information while ensuring trust in the process and integration with scientific workflows. As the role of Artificial Intelligence (AI) continues to grow in the scientific community, the application of AI to scientific data reduction provides further potential opportunities.

Data reduction methods are required across diverse scientific disciplines that align with the Department of Energy’s (DOE) mission, including light sources, accelerators, radio astronomy, genomics, and more. These fields tend to produce massive amounts of raw data that is challenging to manage effectively. Advanced Scientific Computing Research (ASCR) is actively seeking norms, techniques, and workflows that can reduce such data, and these techniques should have the potential to be applied across multiple applications.

In January of 2021, a virtual DOE workshop titled "Data Reduction for Science," discussed these challenges and identified four priority research directions (PRDs). These included reliable and efficient algorithms, progressive reduction algorithms for efficient streaming, algorithms to preserve essential data with quantified uncertainty, and the application of these techniques to new systems and use-cases.

The key objective of this Funding Opportunity Announcement (FOA) is to back mathematical and computer science approaches that can address these PRDs. The research under this FOA should go beyond focusing on particular data sets or specific applications. Instead, it should create a deeper understanding that paves the way for future scientific advances. This means the funding isn't aimed at incrementally extending current research in the proposed area but must reflect viable strategies towards possible solutions of critical problems in science data reduction. Projects that explore innovative ideas or develop unconventional approaches will be highly appreciated, especially those involving techniques such as compression, filtering, and feature extraction. The FOA may also give preference to applications that include reduction estimates for more than one scientific application.