Presenter: Karl Lieberherr, Northeastern University, College of Computer and Information Science, Boston, MA.
Abstract:
We present the Specker Challenge Game (SCG) and our experience in using it for innovation in problem solving software and for disseminating computer science and mathematical knowledge. We present the tools we have developed (including web-based administrators and basic players), those we have planned for the future, our experiences with SCG, its benefits and limitations.
For SCG we turn software for problem solving domains (e.g., optimization and decision problems) into agents that can survive on their own in a "real" virtual world of algorithmic agents/scholars with reputations. During a competition, the scholars propose and oppose scientific hypotheses based on experiments they have conducted. The SCG is designed to promote good scholarly behavior in the scholars. Good scholars propose hypotheses that are difficult to oppose, i.e., hypotheses that are "tight" and that are difficult to discount.
Playing SCG(X) for some domain X, has the following benefits: (1) it focuses researchers on a specific topic by defining a hypotheses language, reducing the amount of management effort. (2) it defines a structured collaboration between researchers. The researchers receive frequent feedback on their work from their peers. This reduces the effort needed to receive targeted feedback. (3) it accumulates knowledge in domain X. Without the game, this knowledge is scattered in emails, programs and minds. (4) researchers get motivated to propose and oppose non-trivial hypotheses in order to gain reputation. They like to win and if they lose, they want to find out why. (5) managers get a fair comparison of the skills of their researchers through the competition results.
To use SCG in a particular domain, a suitable challenge language needs to be designed. We will demonstrate two challenge languages for Maximum Boolean Constraint Satisfaction problems (CSP) to illustrate how to design hypotheses languages and how to find strong hypotheses. We use SCG effectively as a web-based teaching tool in our Software Development courses where the students implement software for Maximum Boolean CSP.
More information on SCG is available: http://www.ccs.neu.edu/home/lieber/evergreen/specker/scg-home.html
Joint work with Ahmed Abdelmeged and Bryan Chadwick