Building a Reactive Citizen Science Platform that Adapts to Users and Machines

Despite the successes of Citizen Science as a new mode of scientific discovery, the needs of the next generation of telescopes and data intensive experiments will outstrip the ability of even the largest citizen science community. What is needed is a system that can combine machine and human effort, where machine learning algorithms can improve with examples from humans and humans can be deployed effectively when machines fail.

Such a platform requires the ability to continually model the proficiencies of its classifiers and the characteristics of its data. We will discuss infrastructure that is currently being developed at the Zooniverse to facilitate and assess such a system touching on our approaches to bootstrapping learning algorithms, real time modeling, decision making and a/b split assesment of potential strategies.