Automated Service Composition in MathServe

The MathServe broker can perform automated web service composition in case one service is not sufficient to answer a given query. Our composition approach combines classical AI planning with the PRODIGY system with decision-theoretic reasoning.
We use a decision-theoretic extension of the Golog language to reason about success probabilities of automated theorem proving services. An offline DTGolog interpreter is used to compute an optimal policy which can then be executed by the MathServe broker. Such an optimal policy was used in the evaluation of MathServe at CASC-20. The following figure illustrates our composition approach.

MathServe's service composition method.
Contact:
Last modified: Fri Sep 22 17:00:59 CEST 2006