TURKPF: TURKONTROL as a Particle Filter
by Ethan Petuchowski, supervised by Matthew Lease and Aashish Sheshadri
"Decision-Theoretic Control of Crowd-Sourced Workflows" by Peng Dai, Mausam, and Daniel S. Weld (2010).
TURKONTROL, an algorithm presented in (Dai et al. 2010), uses a POMDP to model and control an iterative workflow for crowdsourced work. Here, TURKONTROL is implemented using a Particle Filter to reduce the computation time & memory necessary to run it.
TURKPF is not fit for use in the real world. One would have to use Machine
Learning on real data to determine appropriate values and functions for all
of the default parameters and equations. Methods for doing this were described
(Dai et al. 2013).
One would also have to connect to the
Amazon Mechanical Turk API. Other than that, TURKPF is a
functioning autonomous agent for allocating tasks to crowdworkers in an
iterative improvement workflow. Iterative improvement is very flexible, for
example it can be used in a Map-Reduce job-flow in either a Map or a Reduce
The reason to use a particle filter in this situation is for its speed advantages. Being able to choose the number of particles to use is a way to name your desired speed. With the model specifics we used, as described in the tables, an action is always chosen in much less than one second, which is insignificant compared to the time it takes for a crowdworker to come across, accept, and complete a task.