Nowadays, the rapid proliferation of data makes it possible to build complex models for many real applications. Such models, however, usually require large amount of labeled data, and the labeling process can be both expensive and tedious for domain experts. To address this problem, researchers have resorted to crowdsourcing to collect labels from non-experts with much less cost. The key challenge here is how to infer the true labels from the large number of noisy labels provided by non-experts. Different from most existing work on crowdsourcing, which ignore the structure information in the labeling data provided by non-experts, in this paper, we propose a novel structured approach based on tensor augmentation and completion. It uses tensor representation for the labeled data, augments it with a ground truth layer, and explores two methods to estimate the ground truth layer via low rank tensor completion. Experimental results on 6 real data sets demonstrate the superior performance of the proposed approach over state-of-the-art techniques.
|Number of pages
|IJCAI International Joint Conference on Artificial Intelligence
|Published - 2016
|25th International Joint Conference on Artificial Intelligence, IJCAI 2016 - New York, United States
Duration: Jul 9 2016 → Jul 15 2016
ASJC Scopus subject areas
- Artificial Intelligence