We present a Gaussian process (GP) approach to regression with multiple
labels but no absolute gold standard. The GP framework
provides a principled non-parametric framework that can automatically
estimate the reliability of individual annotators from data without the
need of prior knowledge. Experimental results show that the proposed
GP multi-annotator model outperforms models that either average the
training data or weigh individually learned single-annotator models.
Code
The Matlab code can be downloaded here. It has been tested with the free software GPstuff v3.2. My code uses a slightly different syntax than GPstuff v3.2, so a few files need to be overwritten.
References
Perry Groot, Adriana Birlutiu, Tom Heskes.
Learning from Multiple Annotators with Gaussian Processes, ICANN, 2011, 159-164.
[PDF]