This $600,000 Project Grant award from the National Science Foundation's (NSF) Computer and Information Science and Engineering (CISE) program (CFDA 47.070) supports research at North Carolina State University (NC State) to advance the efficiency of sampling and optimization techniques for decentralized machine learning. The overarching goal is to develop improved non-Markovian sampling approaches beyond traditional Markov Chain Monte Carlo methods, with the potential to enhance the performance of distributed machine learning algorithms. Specifically, the project aims to:
Explore adaptive, self-repellent random walk sampling techniques to maximize efficiency for interacting nonlinear Markov chains.
Assess the performance implications of using these advanced sampling approaches to drive distributed stochastic approximation algorithms for machine learning and optimization.
Develop an algorithmic framework to speed up stochastic approximation by integrating it with carefully constructed multi-timescale nonlinear Markovian sampling strategies.
This award reflects NSF's interest in advancing foundational computing research with potential for broad impact across a range of multi-disciplinary applications, including high-dimensional inference, graph-based learning, and decentralized optimization. The project is expected to run from October 2024 through September 2027 at NC State's facilities in Raleigh, North Carolina.