Model-Robust Classification in Active Learning

J. López Fidalgo, J. A. Moler Cuiral, D. P. Wiens

We aim to develop a theory of model-robust classification, and a methodology for applying this to large data sets such as arise in machine learning. The general idea is that there is a (large) population of explanatory variables, which can be easily sampled. With probability a(x; t), an item with covariates x belongs to group A and with probability 1 - a(x; t) it belongs to group .B.. We suppose that the determination of the appropriate group, given x, is difficult and expensive, so that the investigator wishes to sample from x in a manner which is more efficient than random sampling (sometimes termed .passive learning.).

Palabras clave: Big data logistic regression optimal experimental design robust models

M. J. Rivas López, R. Martín Martín, I. García-Camacha Gutiérrez

P. Urruchi Mohino, J. López Fidalgo

C. de la Calle Arroyo, J. López-Fidalgo, L. Rodríguez-Aragón