Joint BCAM-UPV/EHU Data Science and Artificial Intelligence seminar: Extending the learning using privileged information paradigm to logistic regression

LInk to the session here

Abstract
Learning using privileged information paradigm is a learning scenario that exploits privileged features, available at training time, but not at prediction, as additional information for training models. Specifically, learning using privileged information paradigm is addressed from the logistic regression perspective. Two new approaches, LRIT+ and LR+, learned using the privileged information paradigm and preserving the interpretability of conventional logistic regression, are proposed. For its development, the parameters of a traditional logistic regression trained with all available features, privileged and regular, are projected onto the parameter space associated to regular features. The projection to obtain the model parameters is performed by the minimization of two different convex loss functions: for LRIT+ classifier the function is governed by logit terms, and for LR+ by posterior probabilities. Experimental results report improvements of our proposals over the performance of traditional logistic regression learned without privileged information.