Joint Learning with both Classification and Regression Models for Age Prediction

Abstract

Age classification and regression are two main approaches to age prediction in social media, and these two approaches have their own characteristics and strength. For instance, the classification model can flexibly utilize distinguished models in machine learning, while the regression model can capture the connections between different ages. In order to exploit the advantages of both age classification and regression models, a novel approach to age prediction is proposed, namely joint learning for age prediction. Specifically, an auxiliary Long-Short Term Memory (LSTM) layer is employed to learn the auxiliary representation from the classification setting, and simultaneously join the auxiliary representation into the main LSTM layer for the age regression setting. In the learning process, the auxiliary classification LSTM model and the main regression LSTM model are jointly learned. Empirical studies demonstrate that our joint learning approach significantly improves the performance of age prediction using either individual classification or regression model.