Executive Summary : | The study aims to address the challenges of dealing with massive datasets in domains like social networks, epidemiological networks, and Edge AI. Current parallel architectures, such as multi-core and accelerator systems, have memory capacity constraints, making existing algorithms inefficient or failing on large inputs. To address this, the project redesigns algorithms and their implementations in the Parallel Heterogeneous External Memory (PHEM) model, which has input on the disk, is parallel, and runs simultaneously on a heterogeneous computing platform. The proposed project focuses on applications such as viral marketing, recommendation systems, link analysis/prediction, infection spread, misinformation spread, and hotspots from these domains. Solutions to these applications rely on efficient graph/matrix computations such as traversals, shortest paths, centrality metrics, $k$-connectivity, graph/matrix decompositions, and matrix-vector, matrix-matrix products. The project also aims to prepare a library of software routines for other researchers working in this area, along with appropriate feature-rich datasets. The investigators have a strong record of collaboration, co-advising students, and producing research papers. They currently manage their research groups with a mix of doctoral, master's, and undergraduate students. The project will help further collaboration with the PI and co-PI group, provide the necessary thrust to scale their research activity, produce manpower trained in advanced techniques, and publish research in leading conferences and journals, creating multiple downstream opportunities for the people involved. |