The generalized loss function using α-likelihood and its learning

摘要

The method of the minimization of square error has been widely used for the learning of input-output systems. In this study, we propose the specialized information divergence concerning to the stochastic modeling for an input-output system. We show the learning algorithm from the minimization of its divergence and we derive the loss function with respect to the gradient of our learning algorithm inversely and mention some properties of the function. In addition, we mention that its divergence measure involves the Tsallis entropy.