CSE 598: Sparse Scientific Computing  
Padma Raghavan 
343K IST, 5-9233 (raghavan@cse.psu.edu)
Credits: 3

First Meeting: Jan 14, Wednesday, 1:30 p.m. 223B IST
Regular Meeting Times:  F: 1:30:4:00 p.m. 223 B IST

This course concerns advanced scientific computing algorithms for
enabling discovery through computing.  It will focus on  sparse
problems from modeling and data mining that can entertain solutions
with near linear time complexity. However, the inherent parallelism
is complex, representing interactions at multiple scales and
granularities.  Topics covered include  simulations on graphs and
networks,  combinatorial and numeric  problems  such  multilevel
methods for partitioning, clustering and linear solution, sparsity
data mining schemes such as support vector machines, and performance
modeling  of sparse computations on multinode, multicore systems.


There will be two class projects (40% of the  total score) forming
the basic building blocks upon which students can develop a final
research project (60% of the total score). Projects will  involve
algorithm development, parallelization and performance modeling and
analysis.  Lecture notes will be provided on-line along with pointers
to resources on the web and recent research papers.

Introduction  to large-scale modeling applications (4 weeks).
Characteristics of  scientific and engineering applications; and
how they affect algorithm design. Examples from graph mining, disease
network modeling and PDE-based models.  Assignment I: Fast prototyping
of sample applications.

Parallel algorithms and their scalable implementations on multinode
systems with multicores.  (4-5 weeks).  Assignment II:  A parallel
algorithm and its implementation.

Performance modeling, analysis  and scalability (4-5 weeks).  Focus
on multi-objective optimization for sparse graph and matrix algorithms
including multilevel partitioners,  linear solvers and support
vector machines.

Research project presentations (1 week).