A Transfer-Learning Approach to Exploit Noisy Information for Classification and Its Application on Sentiment Detection - Technologies and Applications of Artificial Intelligence - page 265

Algorithm 1. Noise-Label Transfer Learning (NLTL)
We show a simple sketch illustration for the relationship between training in-
stances in Fig. 2, where the red and blue areas denote the instances in high-quality and
low-quality domains respectively. The dark areas represent instances belonging to
both domains. However, the features that represent these instances might have differ-
ent granularity, and the labels in low-quality domain might be incorrect. Each
instance can belong to one of the four groups (as shown in Fig. 2), high-quality do-
main with low-quality mapping, high-quality domain without low-quality mapping,
low-quality domain with high-quality mapping, and low-quality domain without high-
quality mapping. Finally, the task to be solved is defined as given D H and
D L , learn an accurate classifier to predict D H .