AAAI Publications, First AAAI Conference on Human Computation and Crowdsourcing

Font Size: 
Towards a Language for Non-Expert Specification of POMDPs for Crowdsourcing
Christopher H. Lin, - Mausam, Daniel S. Weld

Last modified: 2013-11-03


Crowdsourcing requesters are trapped between a rock and a hard place. Typically they specify their crowdsourcing workflows procedurally, but current languages commit them to overly strict and static policies that waste human effort. While optimizing workflows with more sophisticated tools like POMDPs can significantly reduce labor costs, such advanced AI techniques are hard to use and understand. We report on our progress in developing Clowder, a system that provides the user with an adaptive programming language that looks and feels like Lisp, yet abstracts over POMDPs so that non-experts can write POMDPs without knowing anything about them. Such a system frees requesters from needing to resort to suboptimal techniques that use approximate heuristics or hire a planning expert to formally define and solve their problems.


POMDP; planning; reinforcement learning; utility elicitation

Full Text: PDF