Exponential Family Transfer Learning with Application to Text Document Modeling.

Resumen

Transfer learning, as applied in machine learning, transfers knowledge between similar learning tasks with the objective of improving performance. We propose a method based on the prior selection principle that explores the transfer learning paradigm for estimating probability density functions belonging to the exponential family. Experiments on distribution of words over text documents, modeled as multinomial distributions, have shown better results when compared to maximum-likelihood estimation.