The Center for Discrete Mathematics and Theoretical Computer Science (DIMACS) and the National Institute of Statistical Sciences (NISS) are organizing a workshop on the experimental analysis of algorithms (EAA), whose purposes are to:
EAA holds major opportunities for collaboration between computational and statistical scientists. The most exciting and challenging problems will be those requiring advances in both fields. Longer-range goals include identification and creation of funding sources for collaborative EAA research that aims to have impact on both computer science and the statistical sciences, as well as on users of algorithms. The impacts range from informing selection of algorithms for a particular problem instance to formulating hypotheses about algorithmic behavior that lead to new theoretical results. Topics for potential discussion include: estimating asymptotic performance; sampling and experimental designs for multi-parameter algorithms; use of statistical methods to identify the key contributors to running time; design of convergence tests and stopping criteria for iterative methods; comparing results across platforms; and analyzing streaming data.
The workshop will be highly participatory. We aim to have six presentations by computational scientists on problems in EAA, each accompanied by a statistical response and followed by extensive discussion. In addition to the scheduled talks and responses, there will be an open problem session at which any participant may speak and raise issues that may not otherwise be captured by the formal program.
For additional information, please visit www.niss.org/affiliates/eaa200803.
A PDF version of the workshop announcement is here.