CAREER: Enabling Next-Generation Decentralized Learning: Structure, Models, and Methods

Advances in artificial intelligence, machine learning, and data science are driving innovation across key sectors such as healthcare, infrastructure, and environmental management. However, existing learning approaches struggle to handle the complexity and distributed nature of modern data systems. This project addresses this gap by advancing decentralized learning methods, where multiple agents interact locally over a network without needing a central authority.

These methods improve data privacy, reduce communication bottlenecks, and enhance learning performance by leveraging localized interactions. The outcomes of this project will contribute to critical societal challenges, such as improving digital health monitoring and environmental data processing. In parallel, this research is integrated with a comprehensive educational plan to train a diverse group of future researchers, particularly from underrepresented communities.

The project will strengthen the connection between cutting-edge research and inclusive STEM education through mentorship programs, graduate courses, and outreach initiatives.

This project aims to enable next-generation performance in decentralized learning by addressing fundamental challenges related to communication efficiency, data heterogeneity, and algorithmic complexity. The research is structured around three thrusts. The first thrust focuses on designing finite-time aggregation networks to achieve faster and more efficient convergence in decentralized optimization algorithms, overcoming the limitations of traditional asymptotic approaches.

The second thrust develops algorithms that handle non-classical aggregated costs, such as affine or convex compositions, enhancing the applicability of decentralized learning to a broader range of problems. The third thrust explores high-order optimization methods, including approximate cubic regularization and quasi-Newton techniques, to improve convergence rates and reduce communication costs in decentralized systems.

The project’s contributions will advance the theoretical foundations of decentralized learning while offering practical solutions for scalable, efficient, distributed decision-making across fields such as machine learning, sensor networks, and autonomous systems.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.