mixture density network

A mixture density network (MDN) is a type of neural network with the purpose of determining likely joint probability distributions (which are termed mixtures), based upon incomplete data. An example application might be development of some method to diagnose some disease based upon symptoms, though each individual symptom may be caused by numerous diseases. Modeling the disease likelihood based upon Gaussian distributions for each symptom (which provides a more straightforward mathematical technique for handling the joint distributions) is sometimes ineffective. MDN uses neural-network-based machine learning, aiming to find and use hidden clues that solve the problem.