Summary: Machine Learning has experienced unprecedented growth over the last ten years, especially on classical vision and language tasks. Expectations are high that the next decade of research will continue at the same pace, yielding transformative discoveries in nearly all areas of science. In 2020 the NSF funded a new institute: the National AI Institute for Foundations of Machine Learning (IFML), whose mission is to understand the key foundational questions that need to be solved so that machine learning can continue its upward trajectory.
IFML is a multi-organizational effort comprising personnel from UT-Austin, UW, MSR-Redmond and Wichita State University (see ifml.institute). Part of IFML's mission as mandated by the NSF is to function as a “nexus point” for other institutes and centers that may have overlapping interests in machine learning. We propose a three day working group so that core IFML personnel can collaborate with SFI personnel on topics broadly related to the theory of machine learning.
We are believers in the power of foundational research and its importance to both short and long-term innovation in machine learning. Within the current empirical framework, reducing the amount of trial and error using principled heuristics is extremely impactful. New algorithms and analyses have the potential to dramatically reshape the field. For example, the invention of polynomial-time interior-point methods for solving linear programs has influenced nearly every aspect of optimization, an area at the core of modern machine learning systems. Additionally, foundational researchers are now routinely hired by the largest technology companies. Given this tight integration, theoretical research is positioned to have major influence.