Statistical learning and game theory in crowdsourcing: Theory and practice

Nihar Shah, Martin Wainwright and Kannan Ramchandran

Crowdsourcing (that is, collecting data from people) is pervasive in many applications, starting from collecting labeled data for machine learning algorithms to obtaining reviews and ratings for various products. This data is generally extremely noisy and ill-behaved. Our work involves using game theory and statistical learning theory to obtain high-quality data and draw interesting and useful inferences from it ---- our approach is to develop algorithms that are theoretically sound and work well in practice.

Figure 1
Figure 1: Learning from comparative data