DIMACS, the center for Discrete Mathematics and Theoretical Computer Science, NISS, the National Institute of Statistical Sciences, and SAMSI, the Statistical and Applied Mathematical Sciences Institute, are planning to co-sponsor a series of events addressing the subject of Experimental Analysis of Algorithms: Interfaces Between Statistics and Computer Science.
At this one-day Planning Meeting, a small group of participants from both the algorithms and statistics communities will discuss the topic and follow-up events.
The meeting will introduce experts in experimental algorithmics and in relevant statistical methodologies to one another. The broad goals are to identify research topics of common interest to both communities, and to develop plans for research collaborations and events (such as workshops or speaker series or tutorials) that address these topics.
After some short introductory lectures, we expect to move to a brainstorming format to discuss: What types of statistical problems do experimenters face in algorithmic research? What techniques of statistics and data analysis are most suitable for addressing these problems? What are the high-leverage gaps, and where are the entry points to begin to address them? What problems if any do statisticians see with currently typical papers in experimental analysis of algorithms? What kinds of collaborations between computer scientists and statistical scientists are needed and desirable?
Topics to be discussed may include sampling and experimental design for asymptotic analysis; computer experiments; statistical methods for assessing convergence; variance reduction techniques; and modeling the relationship between input parameters and performance.