Speaker Bio

Kamesh Madduri is an associate professor in the College of Engineering at Pennsylvania State University. He received his Ph.D. in Computer Science from Georgia Institute of Technology’s College of Computing and was previously a Luis W. Alvarez postdoctoral fellow at Lawrence Berkeley National Laboratory. Madduri conducts research on the design of new parallel algorithms and software tools for analyzing massive datasets and in support of large computational science simulations. His current research focuses on four topics: algorithms for graph analysis on emerging parallel systems, computational genomics, algorithms for particle simulations in plasma physics, and indexing and query strategies for high-dimensional scientific and transportation data sets. Madduri has published extensively in the area of high-performance computing, co-authoring over 60 peer-reviewed articles. According to Google Scholar, his published work has been cited over 2800 times with an h-index of 27. He is a recipient of the NSF CAREER award (2013), a co-recipient of the best paper award at the 42nd International Conference on Parallel Processing (2013), and was awarded the first Junior Scientist prize by the SIAM Activity Group on Supercomputing (2010).

Description

Graph-theoretic abstractions are at the core of data-intensive problems arising in social and technological network analysis (e.g., identifying implicit online communities, quantifying centrality and influence in interaction networks, web algorithms), systems biology (for instance, interactome analysis, epidemiological studies, disease modeling), and security applications (e.g., detecting anomalous patterns from socio-economic interactions and communication data). Due to their large memory footprint, irregular access patterns, and low degrees of spatial locality, computations on massive graphs pose serious challenges on current parallel machines. In this talk, I will present my research group’s recent work on enabling large-scale and high-performance graph analysis. Our parallel implementations on multicore servers and leading supercomputers achieve significant parallel speedup for traversal, connectivity, and centrality problems on graph instances in the order of billions of vertices and edges. I will also describe the parallel algorithms and implementation of two software tools that our group has developed: FASCIA for approximately counting and enumerating network motifs, and PULP for multi-objective graph partitioning.