Sheepdog, Parallel Collaborative Programming-by-Demonstration

Programming-by-demonstration (PBD) [1,2] is a class of techniques typically used for improving end-user productivity by automating repetitive tasks or for constructing graphical applications. PBD is mostly considered to be an end-user-programming tool, where a single author demonstrates the task.

In this paper, we propose a different perspective on the field by introducing the concept of collaborative PBD, in which multiple authors contribute demonstrations to an appropriate learning algorithm that is responsible for combining them and producing a distributable procedure model. We concentrate in particular on parallel collaborative PBD, in which multiple recordings of the same task are collected in parallel and combined simultaneously by the learning algorithm.