Interpretable Mixture Density Estimation by use of Differentiable\n Tree-module

In order to develop reliable services using machine learning, it is important\nto understand the uncertainty of the model outputs. Often the probability\ndistribution that the prediction target follows has a complex shape, and a\nmixture distribution is assumed as a distribution that uncertainty follows.\nSince the output of mixture density estimation is complicated, its\ninterpretability becomes important when considering its use in real services.\nIn this paper, we propose a method for mixture density estimation that utilizes\nan interpretable tree structure. Further, a fast inference procedure based on\ntime-invariant information cache achieves both high speed and interpretability.\n