Unimodal regularisation based on beta distribution for deep ordinal regression

Currently, the use of deep learning for solving ordinal classification problems, where categories follow a natural order, has not received much attention. In this paper, we propose an unimodal regularisation based on beta distribution applied to cross-entropy loss. This encourages labels be soft distribution, more appropriate problems. Given that two parameters must adjusted, method automatically determine them is proposed. The regularised loss function used train neural network model with scheme in output layer. results obtained are statistically analysed and show combination these methods increases performance Moreover, proposed performs better than other distributions previous works, achieving also reduced computational cost.