Jurica Levatić will have a talk entitled “Semi-Supervised Learning for Structured Output Prediction: Methods and Applications” on Thursday, January 26. at 13. Invited!
Structured output prediction is concerned with predicting structured, rather than scalar values, such as vectors/tuples of multiple classes/variables, hierarchies or sequences. Due to the possible laborious annotation procedure of such data, labeled samples are often scarce, limiting the predictive performance of the traditional supervised machine learning methods. Semi-supervised learning methods, on the other hand, use unlabeled samples in addition to labeled ones in order to overcome this limitation. In this lecture, we will present the extension of the predictive clustering framework towards semi-supervised learning for three structured output prediction tasks: multi-target regression, multi-label classification and hierarchical multi-label classification. The talk will present the methodology of semi-supervised predictive clustering trees and their applications to real-life problems from several domains, including biology, chemistry, ecology, and remote sensing.